00 voturi pozitive00 voturi negative

16 vizualizări609 paginiM. Rahman
Dalhousie University, Canada

Apr 06, 2019

© © All Rights Reserved

M. Rahman
Dalhousie University, Canada

© All Rights Reserved

16 vizualizări

00 voturi pozitive00 voturi negative

M. Rahman
Dalhousie University, Canada

© All Rights Reserved

Sunteți pe pagina 1din 609

IN FLUID

MECHANICS

IX

Editors

TM WITPRESS M. Rahman &

C.A. Brebbia

Advances in Fluid

Mechanics IX

Visit our website for the current list of titles.

www.witpress.com

WITeLibrary

Home of the Transactions of the Wessex Institute.

Papers presented at Advances in Fluid Mechanics VIII are archived

in the WIT eLibrary in volume 74 of WIT Transactions on

Engineering Sciences (ISSN 1743-3533). The WIT eLibrary provides the

international scientific community with immediate and permanent access to individual

papers presented at WIT conferences.

http://library.witpress.com.

www.engbookspdf.com

NINETH INTERNATIONAL CONFERENCE ON

ADVANCES IN FLUID MECHANICS

CONFERENCE CHAIRMEN

M. Rahman

Dalhousie University, Canada

C.A. Brebbia

Wessex Institute of Technology, UK

R. Bourisli

P. Chu

S. D'Alessio

A. Macpherson

F. Morency

T. Moulden

K. Patra

B. Sarler

S. Sinkunas

L. Skerget

V. Tesar

ORGANISED BY

The Wessex Institute of Technology, UK

University of Split, Croatia

SPONSORED BY

WIT Transactions on Engineering Sciences

International Journal of Computational Methods and Experimental

Measurement

www.engbookspdf.com

WIT Transactions

Transactions Editor

Carlos Brebbia

Wessex Institute of Technology

Ashurst Lodge, Ashurst

Southampton SO40 7AA, UK

Editorial Board

B Abersek University of Maribor, Slovenia G Belingardi Politecnico di Torino, Italy

Y N Abousleiman University of Oklahoma, R Belmans Katholieke Universiteit Leuven,

USA Belgium

P L Aguilar University of Extremadura, Spain C D Bertram The University of New South

K S Al Jabri Sultan Qaboos University, Oman Wales, Australia

E Alarcon Universidad Politecnica de Madrid, D E Beskos University of Patras, Greece

Spain S K Bhattacharyya Indian Institute of

A Aldama IMTA, Mexico Technology, India

C Alessandri Universita di Ferrara, Italy E Blums Latvian Academy of Sciences, Latvia

D Almorza Gomar University of Cadiz, Spain J Boarder Cartref Consulting Systems, UK

B Alzahabi Kettering University, USA B Bobee Institut National de la Recherche

Scientifique, Canada

J A C Ambrosio IDMEC, Portugal

H Boileau ESIGEC, France

A M Amer Cairo University, Egypt

J J Bommer Imperial College London, UK

S A Anagnostopoulos University of Patras,

Greece M Bonnet Ecole Polytechnique, France

M Andretta Montecatini, Italy C A Borrego University of Aveiro, Portugal

E Angelino A.R.P.A. Lombardia, Italy A R Bretones University of Granada, Spain

H Antes Technische Universitat Braunschweig, J A Bryant University of Exeter, UK

Germany F-G Buchholz Universitat Gesanthochschule

M A Atherton South Bank University, UK Paderborn, Germany

A G Atkins University of Reading, UK M B Bush The University of Western Australia,

Australia

D Aubry Ecole Centrale de Paris, France

F Butera Politecnico di Milano, Italy

J Augutis Vytautas Magnus University,

Lithuania W Cantwell Liverpool University, UK

H Azegami Toyohashi University of D J Cartwright Bucknell University, USA

Technology, Japan P G Carydis National Technical University of

A F M Azevedo University of Porto, Portugal Athens, Greece

J Baish Bucknell University, USA J J Casares Long Universidad de Santiago de

Compostela, Spain

J M Baldasano Universitat Politecnica de

Catalunya, Spain M A Celia Princeton University, USA

J G Bartzis Institute of Nuclear Technology, A Chakrabarti Indian Institute of Science,

Greece India

S Basbas Aristotle University of Thessaloniki, J-T Chen National Taiwan Ocean University,

Greece Taiwan

A Bejan Duke University, USA A H-D Cheng University of Mississippi, USA

M P Bekakos Democritus University of

Thrace, Greece

J Chilton University of Lincoln, UK

www.engbookspdf.com

C-L Chiu University of Pittsburgh, USA K Dorow Pacific Northwest National

H Choi Kangnung National University, Korea Laboratory, USA

S Clement Transport System Centre, Australia C Dowlen South Bank University, UK

M W Collins Brunel University, UK J P du Plessis University of Stellenbosch,

South Africa

J J Connor Massachusetts Institute of

Technology, USA

R Duffell University of Hertfordshire, UK

M C Constantinou State University of New N A Dumont PUC-Rio, Brazil

York at Buffalo, USA A Ebel University of Cologne, Germany

D E Cormack University of Toronto, Canada E E Edoutos Democritus University of Thrace,

M Costantino Royal Bank of Scotland, UK Greece

K M Elawadly Alexandria University, Egypt

W Czyczula Krakow University of

Technology, Poland K-H Elmer Universitat Hannover, Germany

D Elms University of Canterbury, New Zealand

M da Conceicao Cunha University of

Coimbra, Portugal M E M El-Sayed Kettering University, USA

L Dávid Károly Róbert College, Hungary D M Elsom Oxford Brookes University, UK

F Erdogan Lehigh University, USA

A Davies University of Hertfordshire, UK

D J Evans Nottingham Trent University, UK

M Davis Temple University, USA J W Everett Rowan University, USA

A B de Almeida Instituto Superior Tecnico, M Faghri University of Rhode Island, USA

Portugal R A Falconer Cardiff University, UK

E R de Arantes e Oliveira Instituto Superior M N Fardis University of Patras, Greece

Tecnico, Portugal

P Fedelinski Silesian Technical University,

L De Biase University of Milan, Italy Poland

R de Borst Delft University of Technology, H J S Fernando Arizona State University,

Netherlands USA

G De Mey University of Ghent, Belgium S Finger Carnegie Mellon University, USA

A De Montis Universita di Cagliari, Italy E M M Fonseca Instituto Politécnico de

Bragança, Portugal

A De Naeyer Universiteit Ghent, Belgium

J I Frankel University of Tennessee, USA

W P De Wilde Vrije Universiteit Brussel,

Belgium D M Fraser University of Cape Town, South Africa

D De Wrachien State University of Milan, M J Fritzler University of Calgary, Canada

Italy T Futagami Hiroshima Institute of Technology,

Japan

L Debnath University of Texas-Pan American,

U Gabbert Otto-von-Guericke Universitat

USA

Magdeburg, Germany

G Degrande Katholieke Universiteit Leuven, G Gambolati Universita di Padova, Italy

Belgium

C J Gantes National Technical University of

E del Giudice University of Milan, Italy Athens, Greece

S del Giudice University of Udine, Italy L Gaul Universitat Stuttgart, Germany

G Deplano Universita di Cagliari, Italy A Genco University of Palermo, Italy

N Georgantzis Universitat Jaume I, Spain

I Doltsinis University of Stuttgart, Germany

P Giudici Universita di Pavia, Italy

M Domaszewski Universite de Technologie de

L M C Godinho University of Coimbra,

Belfort-Montbeliard, France

Portugal

J Dominguez University of Seville, Spain

www.engbookspdf.com

F Gomez Universidad Politecnica de Valencia, Y Jaluria Rutgers University, USA

Spain C M Jefferson University of the West of

R Gomez Martin University of Granada, England, UK

Spain M K Jha Morgan State University, USA

D Goulias University of Maryland, USA P R Johnston Griffith University, Australia

K G Goulias Pennsylvania State University, D R H Jones University of Cambridge, UK

USA

N Jones University of Liverpool, UK

F Grandori Politecnico di Milano, Italy

N Jovanovic CSIR, South Africa

W E Grant Texas A & M University,

USA D Kaliampakos National Technical University

of Athens, Greece

S Grilli University of Rhode Island, USA

N Kamiya Nagoya University, Japan

R H J Grimshaw Loughborough University,

UK D L Karabalis University of Patras, Greece

D Gross Technische Hochschule Darmstadt, A Karageorghis University of Cyprus

Germany M Karlsson Linkoping University, Sweden

R Grundmann Technische Universitat T Katayama Doshisha University, Japan

Dresden, Germany K L Katsifarakis Aristotle University of

A Gualtierotti IDHEAP, Switzerland Thessaloniki, Greece

O T Gudmestad University of Stavanger, J T Katsikadelis National Technical

Norway University of Athens, Greece

R C Gupta National University of Singapore, E Kausel Massachusetts Institute of

Singapore Technology, USA

J M Hale University of Newcastle, UK H Kawashima The University of Tokyo, Japan

K Hameyer Katholieke Universiteit Leuven, B A Kazimee Washington State University,

Belgium USA

C Hanke Danish Technical University, S Kim University of Wisconsin-Madison, USA

Denmark D Kirkland Nicholas Grimshaw & Partners

K Hayami University of Toyko, Japan Ltd, UK

Y Hayashi Nagoya University, Japan E Kita Nagoya University, Japan

L Haydock Newage International Limited, UK A S Kobayashi University of Washington,

A H Hendrickx Free University of Brussels, USA

Belgium T Kobayashi University of Tokyo, Japan

C Herman John Hopkins University, USA D Koga Saga University, Japan

I Hideaki Nagoya University, Japan S Kotake University of Tokyo, Japan

D A Hills University of Oxford, UK A N Kounadis National Technical University

W F Huebner Southwest Research Institute, of Athens, Greece

USA W B Kratzig Ruhr Universitat Bochum,

J A C Humphrey Bucknell University, USA Germany

M Y Hussaini Florida State University, USA T Krauthammer Penn State University, USA

W Hutchinson Edith Cowan University, C-H Lai University of Greenwich, UK

Australia M Langseth Norwegian University of Science

T H Hyde University of Nottingham, UK and Technology, Norway

M Iguchi Science University of Tokyo, Japan B S Larsen Technical University of Denmark,

Denmark

D B Ingham University of Leeds, UK

F Lattarulo Politecnico di Bari, Italy

L Int Panis VITO Expertisecentrum IMS,

Belgium A Lebedev Moscow State University, Russia

N Ishikawa National Defence Academy, Japan L J Leon University of Montreal, Canada

J Jaafar UiTm, Malaysia D Lesnic University of Leeds, UK

W Jager Technical University of Dresden, D Lewis Mississippi State University, USA

Germany S lghobashi University of California Irvine,

USA

www.engbookspdf.com

K-C Lin University of New Brunswick, Canada T Miyoshi Kobe University, Japan

A A Liolios Democritus University of Thrace, G Molinari University of Genoa, Italy

Greece T B Moodie University of Alberta, Canada

S Lomov Katholieke Universiteit Leuven, D B Murray Trinity College Dublin, Ireland

Belgium

G Nakhaeizadeh DaimlerChrysler AG,

J W S Longhurst University of the West of Germany

England, UK

M B Neace Mercer University, USA

G Loo The University of Auckland, New

Zealand D Necsulescu University of Ottawa, Canada

J Lourenco Universidade do Minho, Portugal F Neumann University of Vienna, Austria

J E Luco University of California at San Diego, S-I Nishida Saga University, Japan

USA H Nisitani Kyushu Sangyo University, Japan

H Lui State Seismological Bureau Harbin, B Notaros University of Massachusetts, USA

China P O’Donoghue University College Dublin,

C J Lumsden University of Toronto, Canada Ireland

L Lundqvist Division of Transport and R O O’Neill Oak Ridge National Laboratory,

Location Analysis, Sweden USA

T Lyons Murdoch University, Australia M Ohkusu Kyushu University, Japan

Y-W Mai University of Sydney, Australia G Oliveto Universitá di Catania, Italy

M Majowiecki University of Bologna, Italy R Olsen Camp Dresser & McKee Inc., USA

D Malerba Università degli Studi di Bari, Italy E Oñate Universitat Politecnica de Catalunya,

G Manara University of Pisa, Italy Spain

S Mambretti Politecnico di Milano, Italy K Onishi Ibaraki University, Japan

B N Mandal Indian Statistical Institute, India P H Oosthuizen Queens University, Canada

Ü Mander University of Tartu, Estonia E L Ortiz Imperial College London, UK

H A Mang Technische Universitat Wien, E Outa Waseda University, Japan

Austria A S Papageorgiou Rensselaer Polytechnic

G D Manolis Aristotle University of Institute, USA

Thessaloniki, Greece J Park Seoul National University, Korea

W J Mansur COPPE/UFRJ, Brazil G Passerini Universita delle Marche, Italy

N Marchettini University of Siena, Italy F Patania University of Catania, Italy

J D M Marsh Griffith University, Australia B C Patten University of Georgia, USA

J F Martin-Duque Universidad Complutense, G Pelosi University of Florence, Italy

Spain G G Penelis Aristotle University of

T Matsui Nagoya University, Japan Thessaloniki, Greece

G Mattrisch DaimlerChrysler AG, Germany W Perrie Bedford Institute of Oceanography,

F M Mazzolani University of Naples Canada

“Federico II”, Italy R Pietrabissa Politecnico di Milano, Italy

K McManis University of New Orleans, USA H Pina Instituto Superior Tecnico, Portugal

A C Mendes Universidade de Beira Interior, M F Platzer Naval Postgraduate School, USA

Portugal D Poljak University of Split, Croatia

R A Meric Research Institute for Basic V Popov Wessex Institute of Technology, UK

Sciences, Turkey

H Power University of Nottingham, UK

J Mikielewicz Polish Academy of Sciences,

Poland D Prandle Proudman Oceanographic

Laboratory, UK

N Milic-Frayling Microsoft Research Ltd, UK

M Predeleanu University Paris VI, France

R A W Mines University of Liverpool, UK

I S Putra Institute of Technology Bandung,

C A Mitchell University of Sydney, Australia Indonesia

K Miura Kajima Corporation, Japan Y A Pykh Russian Academy of Sciences,

A Miyamoto Yamaguchi University, Japan Russia

www.engbookspdf.com

F Rachidi EMC Group, Switzerland A P S Selvadurai McGill University, Canada

M Rahman Dalhousie University, Canada J J Sendra University of Seville, Spain

K R Rajagopal Texas A & M University, USA J J Sharp Memorial University of

T Rang Tallinn Technical University, Estonia Newfoundland, Canada

J Rao Case Western Reserve University, USA Q Shen Massachusetts Institute of Technology,

USA

J Ravnik University of Maribor, Slovenia

X Shixiong Fudan University, China

A M Reinhorn State University of New York

at Buffalo, USA G C Sih Lehigh University, USA

G Reniers Universiteit Antwerpen, Belgium L C Simoes University of Coimbra, Portugal

A D Rey McGill University, Canada A C Singhal Arizona State University, USA

D N Riahi University of Illinois at Urbana- P Skerget University of Maribor, Slovenia

Champaign, USA J Sladek Slovak Academy of Sciences,

B Ribas Spanish National Centre for Slovakia

Environmental Health, Spain V Sladek Slovak Academy of Sciences,

K Richter Graz University of Technology, Slovakia

Austria A C M Sousa University of New Brunswick,

S Rinaldi Politecnico di Milano, Italy Canada

F Robuste Universitat Politecnica de H Sozer Illinois Institute of Technology, USA

Catalunya, Spain D B Spalding CHAM, UK

J Roddick Flinders University, Australia P D Spanos Rice University, USA

A C Rodrigues Universidade Nova de Lisboa, T Speck Albert-Ludwigs-Universitaet Freiburg,

Portugal Germany

F Rodrigues Poly Institute of Porto, Portugal C C Spyrakos National Technical University

C W Roeder University of Washington, USA of Athens, Greece

J M Roesset Texas A & M University, USA I V Stangeeva St Petersburg University, Russia

W Roetzel Universitaet der Bundeswehr J Stasiek Technical University of Gdansk,

Hamburg, Germany Poland

V Roje University of Split, Croatia G E Swaters University of Alberta, Canada

R Rosset Laboratoire d’Aerologie, France S Syngellakis University of Southampton, UK

J L Rubio Centro de Investigaciones sobre J Szmyd University of Mining and Metallurgy,

Desertificacion, Spain Poland

T J Rudolphi Iowa State University, USA S T Tadano Hokkaido University, Japan

S Russenchuck Magnet Group, Switzerland H Takemiya Okayama University, Japan

H Ryssel Fraunhofer Institut Integrierte I Takewaki Kyoto University, Japan

Schaltungen, Germany C-L Tan Carleton University, Canada

S G Saad American University in Cairo, Egypt E Taniguchi Kyoto University, Japan

M Saiidi University of Nevada-Reno, USA S Tanimura Aichi University of Technology,

R San Jose Technical University of Madrid, Japan

Spain J L Tassoulas University of Texas at Austin,

F J Sanchez-Sesma Instituto Mexicano del USA

Petroleo, Mexico M A P Taylor University of South Australia,

B Sarler Nova Gorica Polytechnic, Slovenia Australia

S A Savidis Technische Universitat Berlin, A Terranova Politecnico di Milano, Italy

Germany A G Tijhuis Technische Universiteit

A Savini Universita de Pavia, Italy Eindhoven, Netherlands

G Schmid Ruhr-Universitat Bochum, Germany T Tirabassi Institute FISBAT-CNR, Italy

R Schmidt RWTH Aachen, Germany S Tkachenko Otto-von-Guericke-University,

Germany

B Scholtes Universitaet of Kassel, Germany

N Tosaka Nihon University, Japan

W Schreiber University of Alabama, USA

www.engbookspdf.com

T Tran-Cong University of Southern H Westphal University of Magdeburg,

Queensland, Australia Germany

R Tremblay Ecole Polytechnique, Canada J R Whiteman Brunel University, UK

I Tsukrov University of New Hampshire, USA T W Wu University of Kentucky, USA

R Turra CINECA Interuniversity Computing Z-Y Yan Peking University, China

Centre, Italy S Yanniotis Agricultural University of Athens,

S G Tushinski Moscow State University, Greece

Russia A Yeh University of Hong Kong, China

J-L Uso Universitat Jaume I, Spain B W Yeigh SUNY Institute of Technology,

E Van den Bulck Katholieke Universiteit USA

Leuven, Belgium J Yoon Old Dominion University, USA

D Van den Poel Ghent University, Belgium K Yoshizato Hiroshima University, Japan

R van der Heijden Radboud University, T X Yu Hong Kong University of Science &

Netherlands Technology, Hong Kong

R van Duin Delft University of Technology, M Zador Technical University of Budapest,

Netherlands Hungary

P Vas University of Aberdeen, UK K Zakrzewski Politechnika Lodzka, Poland

R Verhoeven Ghent University, Belgium M Zamir University of Western Ontario,

A Viguri Universitat Jaume I, Spain Canada

Y Villacampa Esteve Universidad de R Zarnic University of Ljubljana, Slovenia

Alicante, Spain G Zharkova Institute of Theoretical and

F F V Vincent University of Bath, UK Applied Mechanics, Russia

S Walker Imperial College, UK N Zhong Maebashi Institute of Technology,

G Walters University of Exeter, UK Japan

B Weiss University of Vienna, Austria H G Zimmermann Siemens AG, Germany

www.engbookspdf.com

Advances in Fluid

Mechanics IX

Editors

M. Rahman

Dalhousie University, Canada

C.A. Brebbia

Wessex Institute of Technology, UK

www.engbookspdf.com

Editors:

M. Rahman

Dalhousie University, Canada

C.A. Brebbia

Wessex Institute of Technology, UK

Published by

WIT Press

Ashurst Lodge, Ashurst, Southampton, SO40 7AA, UK

Tel: 44 (0) 238 029 3223; Fax: 44 (0) 238 029 2853

E-Mail: witpress@witpress.com

http://www.witpress.com

For USA, Canada and Mexico

WIT Press

25 Bridge Street, Billerica, MA 01821, USA

Tel: 978 667 5841; Fax: 978 667 7582

E-Mail: infousa@witpress.com

http://www.witpress.com

from the British Library.

ISBN: 978-1-84564-600-4

eISBN: 978-1-84564-601-1

ISSN: (print) 1746-4471

ISSN: (on-line) 1743-3533

The text of the papers in this volume were set individually by the authors or

under their supervision. Only minor corrections to the text may have been

carried out by the ublisher.

No responsibility is assumed by the Publisher, the Editors and Authors for any injury and/or

damage to persons or property as a matter of products liability, negligence or otherwise, or

from any use or operation of any methods, products, instructions or ideas contained in the

material herein. The Publisher does not necessarily endorse the ideas held, or views expressed

by the Editors or Authors of the material contained in its publications.

Printed in Great Britain by Lightning Source, UK.

All rights reserved. No part of this publication may be reproduced, stored in a retrieval

system, or transmitted in any form or by any means, electronic, mechanical, photocopying,

recording, or otherwise, without the prior written permission of the Publisher.

www.engbookspdf.com

Preface

The field of fluid mechanics is vast and has numerous and diverse applications

in many areas of applied mathematics and engineering sciences. AFM 2012

is the ninth International Conference on Advances in Fluid Mechanics held

in Split, Croatia in 2012. The conference is reconvened following the success

of all previous meetings. The papers in this book cover a wide range of

topics in the area of advances in fluid mechanics. The meeting was organised

by the Wessex Institute of Technology, UK and the University of Split,

Croatia.

The conference has reconvened every two years since 1996, and originated

a very successful book series on the same topics which has resulted over 65

volumes since then. The first conference in this series took place in New

Orleans, USA (1996); followed by Udine, Italy (1998); Montreal, Canada

(2000); Ghent, Belgium (2002); Lisbon, Portugal (2004); Skiathos, Greece

(2006); the New Forest, UK (2008); and the Algarve, Portugal (2010).

The papers in the book are arranged in the following sections: Computational

methods; Environmental measurements; Hydrodynamics; Fluid Structure

interaction; Multiphase flow; Applications in biology; Electronic components;

Environmental fluid mechanics; Heat and mass transfer; and Industrial

applications.

in paper and digital format and distributed throughout the world. In addition,

all papers are archived within Wessex Institute electronic library (http://

library.witpress.com) where they are permanently and easily available to the

world scientific community.

www.engbookspdf.com

The Editors are indeed indebted to all authors as well as the members of

International Scientific Advisory Committee who helped review the papers.

The Editors,

Split, Croatia, 2012

www.engbookspdf.com

Contents

M. Rahman, S. H. Mousavizadegan & D. Bhatta ................................................ 3

G. V. Kudav, Y. M. Panta & M. Yatsco ............................................................. 15

A. Goharzadeh, L. Khezzar & A. Molki ............................................................. 29

F. Concli & C. Gorla ......................................................................................... 37

R. A. Saeed, V. Popov & A. N. Galybin ............................................................. 49

Y. Y. Pey, L. P. Chua & W. L. Siauw ................................................................. 59

T. H. Moulden .................................................................................................... 73

Z. Qian & C.-H. Lee .......................................................................................... 85

viscous flow in complex geometries

B. Jastrow & F. Magagnato .............................................................................. 97

J. Skibinski, T. Wejrzanowski, J. Szumbarski & K. J. Kurzydlowski ............... 109

www.engbookspdf.com

Assessment of two pressure-velocity coupling strategies for local meshless

numerical method

G. Kosec, R. Vertnik & B. Šarler ..................................................................... 119

E. Eslamian, H. Shirvani & A. Shirvani .......................................................... 131

S. Djellal, T. Azzam, M. Djellab & K. Lakkaichi ............................................. 141

in an axial flow

Z. G. Liu, Y. Liu & J. Lu .................................................................................. 151

using the spectral element method

H. Gun Sung, H. Baek, S. Hong & J.-S. Choi .................................................. 163

aeronautics

O. Gohardani & D. W. Hammond ................................................................... 175

evaporating water droplets and vapour condensation intensity

G. Miliauskas, S. Sinkunas & K. Sinkunas ...................................................... 189

with in-situ microscopy

T. Krebs, J. J. Slot, C. P. G. H. Schroen,

H. W. M. Hoeijmakers & R. M. Boom ............................................................. 199

S. Martel, Y. Mercadier & M. Dostie............................................................... 211

vertical pipe with phase change using a multi-fluid modelling approach

R. Kopun & L. Škerget ..................................................................................... 219

fluidized bed reactors

R. K. Thapa & B. M. Halvorsen....................................................................... 231

www.engbookspdf.com

Study of a fluidized bed reactor for gasification of biomass

K. Kleiva, R. K. Thapa & B. M. Halvorsen...................................................... 241

A. Gajewski ...................................................................................................... 251

R. Denefle, S. Mimouni, J.-P. Caltagirone & S. Vincent ................................. 263

approach by way of statistical bubble micro-flow modelling

W. Coetzee , R. L. J. Coetzer & R. Rawatlal .................................................... 275

in an industrial planetary speed reducer

F. Concli & C. Gorla ....................................................................................... 287

Water and contaminant flux estimation from multi-layer passive flux meter

measurements

H. Klammler, K. Hatfield, J. Luz, M. Annable, M. Newman, J. Cho,

A. Peacock, V. Stucker, J. Ranville & C. Clark II............................................ 301

geometry

G. E. Swaters ................................................................................................... 313

M. Bencivenga, G. Nardone, F. Ruggiero & D. Calore .................................. 321

PIV measurements on the formation of the flow field and aerosol particle

distribution in a turbulent square duct flow

T. Barth, M. Banowski & U. Hampel............................................................... 335

discrete particles

K. Kumaran & R. Sadr .................................................................................... 345

over dynamic rippled beds

A. M. Penko & J. Calantoni ............................................................................. 361

www.engbookspdf.com

Validation of a stochastic droplet breakup model applied to a liquid jet in

turbulent crossflow

R. Blanchard & S. Shi ...................................................................................... 371

computational fluids dynamics (CFD)

F. Vásquez, M. Stanko, A. Vásquez, J. De Andrade

& M. Asuaje ..................................................................................................... 381

two-phase fluids

J. V. Madison ................................................................................................... 393

on-line monitoring of bubble phenomena and flow patterns in two-phase gas-

liquid flow

S. Husin, A. Addali & D. Mba ......................................................................... 403

S. Husin, A. Addali & D. Mba ......................................................................... 415

T. Krebs, C. P. G. H. Schroen & R. M. Boom.................................................. 427

solidification process of water in an annular enclosure

E. M. Alawadhi & R. I. Bourisli....................................................................... 441

S. J. D. D’Alessio & K. A. Ogden .................................................................... 453

in a deepwater wellbore

B. J. Sun, Y. H. Gao, Y. Q. Ma, Z. Y. Wang & H. Li ....................................... 465

variable viscosity

A. Fatsis, J. Statharas, A. Panoutsopoulou & N. Vlachakis ............................ 481

J. Kordík & Z. Trávníček ................................................................................. 493

www.engbookspdf.com

Section 8: Hydrodynamics

V. Tesař............................................................................................................ 507

(east Adriatic)

G. Lončar, G. Beg Paklar & V. Andročec ....................................................... 521

Z.-H. Silber-Li, X. Zheng, G.-P. Kong & T. H. Moulden ................................. 533

R. W. Derksen .................................................................................................. 547

A. K. Macpherson, S. Neti, M. Averbach, P. A. Macpherson

& C. Chutakositkanon ..................................................................................... 559

F. Lizal, M. Fusek, J. Jedelsky & M. Jicha ...................................................... 571

www.engbookspdf.com

This page intentionally left blank

www.engbookspdf.com

Section 1

Fluid structure interaction

www.engbookspdf.com

This page intentionally left blank

www.engbookspdf.com

Advances in Fluid Mechanics IX 3

a classical perturbation method

M. Rahman1 , S. H. Mousavizadegan2 & D. Bhatta3

1

Faculty of Computer Science, Dalhousie University, Halifax, Canada

2

Faculty of Marine Technology, Amirkabir University of Technology, Iran

3

The University of Texas Pan-American, Edinburg, USA

Abstract

This paper deals with the classical theory of the perturbation method to determine

the solution of wave forces on a circular cylinder in regular waves. Nonlinear

diffraction of water waves is considered here to demonstrate the powerful

perturbation technique. The solution obtained by this method is compared with

the available experimental data. It is found that the analytical solution agrees with

the experimental data quite well. When we talk about wave effects on offshore

structures, it is necessary to distinguish between small and large dimensions in

relation to the characteristic wavelength and the wave amplitude. It is well known

that the Morison equation displayed an empirical relationship in terms of the

coefficient of mass, CM , and of the coefficient of drag, CD , the two hydrodynamic

coefficients used to calculate wave forces on a small submerged cylinder. This

relationship involves the inertia force and viscous drag force on the cylinder and

assumes that the object is small so as not to disturb the incident wave field.

However, as the diameter of the cylinder becomes large compared to the incident

wavelength, the Morison equation does not apply and a diffraction theory must

be used. In this case, viscous drag forces are assumed to be insignificant for

smooth dimension structures (cylinders) and the inertia forces predominate. In this

paper we discuss the theoretical formulation of second order wave loads. We have

obtained the mathematical expressions for the forces and moments to predict the

wave loads on large monolithic offshore structures.

Keywords: Morison equation, inertia force, drag force, nonlinear diffraction,

radiation, wave forces, moments, perturbation method, second-order theory,

offshore structures, cylinders, diffracted waves, incident waves.

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120011

www.engbookspdf.com

4 Advances in Fluid Mechanics IX

1 Introduction

The determination of wave forces on offshore structures is essential to study

the effects of waves, wind and current on them. The offshore structures should

experience minimal movement to provide a stable work station for operations

such as drilling and production of oil. The solution to the problem of ocean wave

interaction with offshore structures is usually very complex. In many cases, only

an approximate solution is sought. Some of the mathematical techniques required

for the hydrodynamic problem associated with the design of offshore structures are

analytical while many are numerical in nature. While the evolution of computers

has made the numerical methods more advantageous over the classical analytical

methods, numerical methods alone cannot find absolute success without being

complemented by either analytical methods or at least experiments; in this sense

analytical methods become a cost efficient and handy technique for designers in

most cases.

Morison et al. [1] gave an empirical relationship in terms of coefficient of mass,

CM , and of coefficient of drag, CD , the two hydrodynamic coefficients to calculate

wave forces on a small submerged cylinder in regular waves. However, when the

size of the structure is large compared to the wavelength, there are considerable

scattering and diffraction in the wave pattern and in that situation the Morison

equation is not suitable to calculate the wave loads and we must use the diffraction

theory. A linear diffraction theory was first formulated by MacCamy and Fuchs

[2] to calculate the wave loads on a vertical circular cylinder. The cylinder extends

from well above the surface to the bottom of the sea. A number of investigators

including Mogridge and Jamieson [3] have used this theory to obtain the wave

loads on large submerged cylinders in the sea. In all these investigations, the

analyses were restricted to linear theory only, which has very limited applications

in a practical field because the character of the ocean waves is very often nonlinear.

To improve the correlation between experimental data and theory, it is necessary

to include the nonlinear effects in the diffraction theory.

Taking into consideration this motivation, many investigators have directed their

attention towards nonlinear theory in water waves. Chakrabarti [4] obtained an

expression for wave forces on a cylinder using Stokes’s fifth order theory, but

the kinematic boundary condition was not satisfied in the vicinity of the cylinder.

The results obtained were compared with the second order theory of Yamaguchi

and Tsuchiya [5] and he concluded that his theory and experimental showed

good agreement with their experimental data. Raman and Venkatanarasaiah [6]

obtained a nonlinear solution using a perturbation technique. Unfortunately, their

experimental results were obtained using a relatively small diameter cylinder, for

which the previous workers had assumed the viscous drag forces to be significant.

This paper deals with the theory of second-order wave load expressions to

predict the wave forces and moments on large monolithic offshore structures.

The theory is applied to a surface-piercing cylinder in regular waves. Following

Stoker [7], we have developed a mathematical model of the nonlinear wave theory

to correlate with nature. A perturbation technique has been used to solve the

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 5

nonlinear diffraction problem. For the case of fixed offshore structures, closed

form solutions for drift forces and moments have been obtained using the second-

order theory, and the predictions have been compared with available experimental

data.

2 Mathematical formulation

A rigid vertical surface-piercing circular cylinder of radius a and the diameter

D = 2a is acted upon by a train of regular surface waves of height H, progressing

in the positive x− direction (Figure 1). It is assumed that the fluid is incompressible

and the motion irrotational. When the amplitude is large, the small amplitude

theory does not hold good. In practice, the finite amplitude wave theory, namely

the nonlinear wave theory, is of primary importance. In linear wave, the wave

amplitudes to the second and higher orders are considered negligible, whereas in

the finite amplitude wave theory these higher order terms are retained so as to give

an accurate representation of the wave motion.

Let φ(r, θ, z, t) denote the total fluid velocity potential and let z = η(r, θ, t) be

the equation of the free surface, where (r, θ, z) are the cylindrical coordinates.

Then everywhere in the region of the flow, the fluid motion is governed by

Laplace’s equation.

eta x

y

theta b h

waves.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

6 Advances in Fluid Mechanics IX

∂ 2 φ 1 ∂φ 1 ∂2φ ∂2φ

∇2 φ = 2

+ + 2 2 + 2 = 0, (1)

∂r r ∂r r ∂θ ∂z

in the region

a ≤ r < ∞; −h ≤ z < η; −π ≤ θ ≤ π.

Here ∇ is the Laplacian operator. Equation (1) is Laplace’s equation and it is

derived from the fluid continuity equation. The boundary conditions associated

with the equation (1) are as follows:

Bottom boundary condition:

Assuming the floor of the ocean to be flat, the boundary condition at the ocean

bottom states that the vertical component of the velocity is zero at the bottom

∂φ

= 0 on z = −h; (2)

∂z

Body surface boundary condition:

On the surface of the body the velocity of the fluid must be equal to zero

∂φ

= 0 on r = a; −h ≤ z ≤ η (3)

∂r

Dynamic free surface boundary condition:

The dynamic free surface condition is derived from the Bernoulli equation, on

the assumption that the atmospheric pressure outside the fluid is constant.

∂φ 1 ∂φ 2 1 ∂φ 2 ∂φ

+ gη + ( ) +( ) + ( )2 = 0; z = η, r ≥ a. (4)

∂t 2 ∂r r ∂θ ∂z

Kinematic free surface boundary condition:

The kinematic condition states that a particle lying on the free surface will

continue to remain on the surface. Mathematically,

∂η ∂φ ∂η 1 ∂φ ∂η ∂φ

+ ( )( ) + 2 ( )( ) = ; on z = η, r ≥ a. (5)

∂t ∂r ∂r r ∂θ ∂θ ∂z

In addition to the above boundary conditions, there is the Orr-Sommerfeld

radiation

condition to be satisfied by the scattering potential Φs , where φ =

Re (ΦI + Φs )e−iσt in which ΦI and Φs are defined to be the complex incident

and complex scattered potentials respectively. This boundary condition states that

the diffracted wave must vanish at infinity and mathematically can be defined by

the Orr-Sommerfeld condition

√ ∂

lim r ± ik Φs = 0, (6)

r→∞ ∂r

√

where −1 = i and k is the incident wave number. Here Re stands for the real

part, σ is the wave frequency.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 7

∂φ 1 ∂φ ∂φ

vr = , vθ = , and vz = ,

∂r r ∂θ ∂z

where φ is the total velocity potential, η is the height of the free surface, h is

the depth of water below the still water level, vr , vθ and vz are the velocity

components, g is the acceleration due to gravity, k is the incident wave number (=

2π/L), x, y, z are rectangular coordinates, r, θ, z are the cylindrical coordinates,

t is the time variable and L is the wavelength. The Cartesian form of (1) is

∇2 φ = + 2 + 2 =0 (7)

∂x2 ∂y ∂z

The formulas for the forces F and moments M of the fluid on the body S can

be written in vector form as follows:

F n

= P dS,

M S r×n

where n is the outward normal vector from the body surface, P is the pressure

field which can be obtained from the Bernoulli’s equation as illustrated in the next

section and r is the radial distance vector.

The problem is to work with the complete form of the equation to assume that the

solution can be represented in terms of a power series expansion of the parameter

ε where ε = kH/2 = kA. A is the amplitude of the wave. Thus expanding φ, η

and P as a series in powers of ε yields

φ= εn φn , (8)

n=1

∞

η= εn ηn , (9)

n=1

∞

P = εn Pn . (10)

n=0

The sum of the terms up to index n represents the nth-order theory for any

particular quantity, P represents the pressure field, and φ, η and P are all functions

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

8 Advances in Fluid Mechanics IX

φ(x, y, z, t) = φ[x, y, η(x, y, t), t]. (11)

The modified velocity potential at the free surface is

∂φ1

φ(x, y, η, t) = εφ1 (x, y, 0, t) + ε2 [φ2 (x, y, 0, t) + η1 ( )z=0 ] + O(ε3 ).

∂z

(12)

Substituting the series into the dynamic and kinematic free surface boundary

conditions, it is found that the first-order potential φ1 and the second-order

potential φ2 satisfy the following equations, respectively:

∂ 2 φ1 ∂φ1

+g = 0 on z = 0, (13)

∂t2 ∂z

∂ 2 φ2 ∂φ2 ∂ ∂ 2 φ1 ∂φ1

+ g = −η1 + g

∂t2 ∂z ∂z ∂t2 ∂z

∂ ∂φ1 2 ∂φ1 2 ∂φ1 2

− ( ) +( ) +( ) , r ≥ a. (14)

∂t ∂x ∂y ∂z

The pressure P (r, θ, z, t) may be determined from the Bernoulli’s equation and

on substitution for φ as a series in power of ε, leads to

∂φ1 ∂φ2 1 ∂φ1 2

P = −ρgz − ερ − ε2 ρ[ + (( )

∂t ∂t 2 ∂r

∂φ1 2 1 ∂φ1 2

+ ) + 2 ) )] + O(ε3 ). (15)

∂z r ∂θ

The total horizontal force is

2π η

Fx = |P |r=a (−a cos θ)dzdθ. (16)

0 −h

Fx = εFx1 + ε2 Fx2 , (17)

where the first order contribution is

2π 0

∂φ1

εFx1 = aρ (ε )r=a cos θdθ (18)

0 −h ∂t

and the second-order contribution is

2π εη

2 ∂φ1

ε Fx2 = aρ [ (gz + ε )r=a dz

0 0 ∂t

0

∂φ2 1 ∂φ1 2 1 ∂φ1 2

+ ε2 + ( ) + 2( ) dz] cos θdθ.

−h ∂t 2 ∂z 2a ∂θ r=a

(19)

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 9

It is to be noted here that in the second-order force evaluation given in (19), the

first term on the right can be defined as the waterline force, the third term as the

dynamic force and the second term as the quadratic force as stated by Lighthill [8].

The expressions for φ1 and φ2 are given as follows (Rahman [9]):

∞

σ cosh k(z + h) −iσt

k 2 sinh kh m=0

−2iσt

∞ ∞

σe

φ2 = Dm (k2 )Am (k2 r) cosh k2 (z + h)dk2 cos mθ.

2k 2 m=0 0

Using the results presented by Rahman and Heaps [10], the non-dimensional forms

of the first-order and second-order components of the total forces may now be

written as

Fx1 tanh kh cos(σt − α1 )

= . , (20)

ρgD 3 2(ka)3 |H (1) (ka)|

1

experimental data of Chakrabarti [4].

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

10 Advances in Fluid Mechanics IX

Figure 3: Potential drift force on fixed circular cylinders, h/a = 1.16: Comparison

with the numerical data of Garrison [11].

Comparison with the numerical data of Garrison [11].

and −2iσt

Fx2 e tanh kh ∞

= G(k2 )dk2 + c.c.

ρgD3 8(ka) 0

∞

1 2kh ( + 1)

− 4

(−1) [(3 − )+

4π(ka) sinh 2kh a2 k 2

=0

2kh

× (1 + )] × [C cos 2σt − S sin 2σt]

sinh 2kh

∞

1 ( + 1) 2kh

− [(1 − )(1 + )E ] (21)

4π(ka)4 a2 k 2 sinh 2kh

=0

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 11

cylinders (Experimental data of Mogridge and Jamieson [3]).

where ∞

ka A1 (k2 r)B1 (kr)d(kr)

G(k2 a) = . (22)

(1)

(k2 a) (k2 a) − 4ka tanh kh

tanh k2 h H1 (k2 a)

The second-order force (21) contains two parts; one oscillatory and the other

steady state. They are defined, respectively, as

Fx2 F OS F SS

3

= 2 3 + 2 3.

ρgD ρgD ρgD

moments are given by

εM1 (H/L)π

= CM [kh tanh kh + sechkh − 1] cos(σt − α1 ), (23)

ρgD4 16(D/L)(ka)

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

12 Advances in Fluid Mechanics IX

and

∞

ε2 M 2 (H/L)2 (h/D)

( + 1)

= [ 2 2 B − C]E

ρgD4 4π 3 (D/L)4 k a

=0

∞

(H/L)2 (h/D) −2iσt

+[ tanh khe G(k2 a)λ(k2 h)d(k2 a) + c.c.]

8(D/L)2 0

∞

(H/L)2 (h/D)

( + 1)

− 3 4

(−1)+1 [A + B]

4π (D/L) k 2 a2

=0

The second-order moment (24) contains two parts; one oscillatory and the other

steady state. They are defined, respectively, as

M2 M2OS M2SS

= + .

ρgD4 ρgD4 ρgD4

The graphical representations of these results are displayed with the comparison

of the experimental and numerical data. It is worth mentioning here that all the

parameters defined here are found in the work of Rahman [9].

The analytical and numerical and experimental results are displayed and discussed

in this section. In Fig. 2, both the first-order and the second-order solutions (see

Rahman [9]) are compared with the force measurements of Chakrabarti [4], which

are generally seen to be closer to the second-order predictions. The normalized

dynamic drift forces and moments are potted against the diffraction parameter

ka in Fig. 3 and Fig. 4 using the analytical results. This results are compared

with the numerical solutions obtained by Garrison [11] and the present predicted

results quite agree with the Garrison’s numerical data. Garrison used the Green

function method to determine the drift forces and moments. In Fig. 5, analytical

solutions are compared with the experimental data of Mogridge and Jamieson [3]

for a vertical circular cylindrical structure. These results were computed taking the

averages of the positive and negative maximum loads. The theoretical predictions

of moments seem to agree very well with the measured values for the given range

of parameters.

Acknowledgements

We are grateful to the Natural Sciences and Engineering Research Council of

Canada for its financial support leading to this paper. We are also thankful to the

Faculty of Computer Science at Dalhousie University for the computer facilities

provided to complete this paper.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 13

References

[1] Morison, J. R., O’Brien, M. P., Johnson, J. W. and Schaaf, S. A. (1950). The

forces exerted by surface waves on piles, J. Petrol. Technology, 186, 149–

154.

[2] MacCamy, R. C. and Fuchs, R. A. (1954). Wave forces on piles: a diffraction

theory, U. S. Army Beach Erosion Board, Techn. Memo. No. 69, December,

17 pages.

[3] Mogridge, G. R. and Jamieson, W. W. (1975). Wave forces on circular

caissons: theory and experiment, Can. J. Civ. Engg., 2, 540–548.

[4] Chakrabarti, S. K. (1975). Second-order wave forces on large vertical

cylinders, Proc. ASCE, 101, WW3, 311–317.

[5] Yamaguchi, M and Tsuchiya, Y. (1974). Nonlinear effects of wave pressures

and wave forces on a large cylindrical pile, Proc. Civ. Engg. Soc. in Japan,

229, 41–53 (in Japanese).

[6] Raman, H. and Venkatanarasaiah, P. (1976). Forces due to nonlinear water

waves on vertical cylinders, Proc. ASCE, 102, WW3, 301–316.

[7] Stoker, J. J. (1957). Water waves, Interscience, New York.

[8] Lighthill, M. J. (1979). Waves and hydrodynamic loading, Proc. Second Int.

Conf. on The Behaviour of Offshore Structures, BOSS’79, London, 1, 1–40.

[9] Rahman, M. (1988). The hydrodynamics of waves and tides with

applications, Topics in Engineering Vol. 4. Edited by Carlos Brebbia and

J. J. Connor, Computational Mechanics Publications, Southampton UK and

Boston USA.

[10] Rahman, M. and Heaps, H. S. (1983). Wave forces on offshore structures:

nonlinear wave diffraction by large cylinders, J. Physical Oceano., 13, 2225–

2235.

[11] Garrison, C. J. (1984). Wave structure interaction, Proc. Specialty Conf.

Comput. Methods in Offshore Engg., CSCE, Halifax, N.S., Canada, 1–72.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

This page intentionally left blank

www.engbookspdf.com

Advances in Fluid Mechanics IX 15

roof-mounted solar panels

G. V. Kudav1, Y. M. Panta1 & M. Yatsco2

1

Youngstown State University, Department of Mechanical Engineering,

Ohio, USA

2

V&M Star, Youngstown, USA

Abstract

The amount of solar energy that can be produced in the US and throughout the

world has seen an unprecedented potential to fulfill growing energy demand.

Solar panels can be installed on the ground or on the roof of a building. Roof

mounted solar panels could experience occasional high wind loads especially lift

and drag forces. Solar panels are bolted directly onto the roof and are secured

using ballasts as counter weights against the wind loads. We propose the use of

efficient wind deflectors designed and strategically placed in front of the panels

as reported here. The deflectors under study were proven to minimize the wind

loads on solar panels, ensuring the safety of civilians and surrounding property.

The present study utilizes wind tunnel testing and computational simulation

using the commercial computational fluid dynamics software ANSYS Fluent for

a steady, turbulent wind flow (standard k-ε model) over an inclined rooftop-solar

panel. The study shows promising results in the prediction of the wind forces for

an effectively designed wind management system. As specified earlier, results

were compared and validated by both wind tunnel experiments and

computational simulations for a meaningful conclusion. Solar panels with

various aspect ratios for high incoming wind speeds in the range 40–50 m/s (i.e.

90–110 mph) with several angles of attack were modeled and simulated. We

report the analysis of high wind loads on the solar panels leading to the design of

an optimized wind deflector to counter such loads. It was concluded that an

elliptically profiled wind deflector, with uniformly spaced short fins that were

positioned before the tilted panels, was proven to minimize the high wind loads

by as much as half, compared to the wind loads without the deflector.

Keywords: rooftop solar panels, solar panel deflectors, wind loads, ballast.

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120021

www.engbookspdf.com

16 Advances in Fluid Mechanics IX

1 Introduction

Solar power has gained several increasing domestic and commercial applications

as a renewable source of energy [1]. The two most common methods of solar

panel deployment are on the ground or on the roof. Ground-mounted solar panel

racks need open land that can considerably increase initial project investment

and also the system installation costs. The more logical and cost saving

alternative would be to mount the solar panel racks on the roof of a building

without the need to purchase or lease an open lot to install the solar panel

system. This also enables the solar panels to be installed even in densely

populated locations where space costs are at a premium. Other advantages of

installing the racks on the roof are the ease of installation and maintenance in

that there is no need to dig holes for the support beams of the solar panel racks as

there is for ground-mounted racks.

Earlier research studies indicate that the optimum tilt angles for the solar

panels are between 22o and 48o which can allow sufficient sunlight in solar

panels. Tilt angles close to about 30o are the best for various regions in the US as

indicated by figure 1. One previous study [2] has presented some good strategies

for mounting solar panel arrays on roofs. It is important to mention here that

these tilt angles are influenced by the latitude of the location and the solar

azimuth angle. A recent study [3] reported that wind tunnel experiments can

successfully analyze wind uplift on solar panel models. The study also discussed

the effect of guide plates for the reduction of lift forces in various experimental

configurations of the solar panels. A more recent study revealed the usefulness of

wind tunnel experiments and computational fluid dynamics (CFD) analysis for

roof-mounted solar photovoltaic arrays in correlating CFD simulation results

with the wind tunnel test data [4].

Figure 1: The solar azimuth angle for a photovoltaic array situated in the

Northern hemisphere [5].

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 17

The solar panel racks can be installed on almost any type of flat roof surface

made of tar, gravel, or rubber. Major considerations for the roof-mounted panel

racks are wind speed and wind direction resulting in aerodynamics forces or

wind loads on the racks. General wind data maps [2] help determine the ranges

of wind speeds, their direction, and the angle of attacks in the continental United

States throughout the year. Similarly, general wind data maps of other continents

and countries can also be used to estimate the ranges of wind speeds and their

direction/angle of attacks. Wind loads often necessitate the use of heavy ballast

to counter such loads. The ballast weight, however, could exceed the allowable

structural load on the roof. Therefore, efficient wind deflecting systems could

reduce or ideally eliminate use of ballast.

A recent study revealed the usefulness of wind tunnel experiments and CFD

analysis for roof-mounted solar photovoltaic arrays [6]. With the aid of CFD and

wind tunnel facilities, the current work presents the design of wind deflectors to

minimize the wind loads, especially the drag and lift forces, on the solar panels.

The computational analysis comprises the use of CFD software ANSYS Fluent

in an iterative mode for one-quarter and full scale models. These results were

further validated with wind tunnel tests. Several geometries of wind deflectors

were designed and analyzed to minimize wind load reduction and the one that

reduced the overall drag and lift forces the most was ultimately chosen and

offered to the research sponsor – Northern States Metal, NSM, Youngstown,

Ohio, USA [7] – for possible development and marketing.

2 Theory

An algorithm with a set of governing equations including i) the mass/continuity

(eqn (1)), ii) the momentum/Navier-Stokes (eqn (2)), iii) the turbulent kinetic

energy (eqn (3)), and iv) the dissipation rate equations (eqn (4)) are used for

mathematical modeling of the wind flow domain around the solar panel. These

governing equations are embedded within ANSYS Fluent to analyze the fluid

properties, specifically the lift and drag characteristics on (i) the solar panel and

(ii) the solar panel with the deflector placed in front of the panel. Mathematical

modeling was carried out in order to analyze the wind forces of the incoming

high-speed wind on the given panel model and the deflector. The standard k-ε

turbulence model for the turbulent flow was chosen for ANSYS Fluent [8, 9].

(1)

Momentum – Navier-Stokes Equations:

(2)

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

18 Advances in Fluid Mechanics IX

(3)

(4)

(5)

where the nomenclature for the symbols has the usual meaning and they are

defined below [7]:

u , v, w x-, y-, and z-components of the velocity, respectively

f = body forces

k = turbulent kinetic energy

ε = the rate of dissipation.

µt = the turbulent viscosity

Gk generation of turbulent kinetic energy due to the mean velocity gradient

Gb generation of turbulent kinetic energy due to buoyancy

YM contribution of the fluctuating dilatation in compressible turbulence

overall dissipation rate

S k user defined source term

S user defined source term

Other arbitrary constants in Eqns. (3) and (4) are used from the ANSYS

Fluent Manual: C1ε = 1.44, C2ε = 1.92, Cµ = 0.09, σk = 1.00, and σε = 1.3. These

values are acceptable for wall-bounded and free shear flows and are appropriate

for a CFD study of solar panel racks positioned in a wall-bounded fluid domain

[8, 9].

3 Methodology

As ballast for the solar panel to add substantial weight on a roof mounting, a

wind deflector was conceptualized as an alternative solution to reduce or replace

the use of such ballast. Based on the study, it was recommended that wind

deflectors were placed on all sides of the solar panel racks as a complete wind

management system to reduce the head-wind and side-wind effects. If all sided

wind deflectors are not economically viable, the wind deflector on the front is

highly recommended to minimize the head-wind effects. Although the wind

deflector itself will add a minimal weight on the roof it is negligible compared to

that of ballast. Multiple deflectors of various surface profiles were designed that

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 19

included vertical, inclined, parabolic, elliptic and elliptic with fins shapes. Only a

sketch of the elliptic deflectors is shown in figure 2.

Wind flows over the physical model with and without a deflector were tested

for wind forces using wind tunnel instruments. Similarly, wind flows with and

without a deflector were simulated, and analyzed for wind forces using ANSYS

Fluent software. Based on the preliminary CFD analysis among the deflector

shapes for high winds, an elliptic finned-deflector (figure 2(b)) was chosen for

further analysis by CFD simulation and testing in the wind tunnel for its relative

superior performance.

(a) (b)

Figure 2: Side views of the deflector: (a) elliptic, (b) elliptic plus fins.

simulations and experiments using wind tunnel testing were employed for wind

load analysis. These methodologies are given below.

algorithm and mathematical model. ANSYS Fluent software that combines

numerical techniques with the intricacies of fluid flow is employed for the

modeling of wind flow over the solar panel. The software was built to model and

analyze many types of laminar and turbulent fluid flows. The software has

different packages and add-ons that allow one to model various geometries with

the chosen flow models. The software comes with geometry modeling software,

called ANSYS Design Modeler. The remaining add-ons include ANSYS

Meshing, ANSYS Fluent, and CFD Post. Details of the ANSYS Fluent,

computational models/simulations for the wind flow analysis of the rooftop solar

panel racks are discussed in the sections that follow.

The procedure to set-up and run a successful simulation in ANSYS Fluent, for a

fluid flow problem, consists of a series of steps that are completed sequentially.

The procedure is outlined as preprocessing, processing, and post-processing. The

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

20 Advances in Fluid Mechanics IX

Fluent are summarized below:

1. Construction of the geometrical models using ANSYS Design Modeler.

2. Division of the fluid domain into discrete volumes using ANSYS Meshing.

3. Modeling using ANSYS Fluent (figure 3).

4. Defining the boundary conditions and fluid properties.

5. Solving in Fluent until a converged solution is achieved (eqn. (6)).

Update properties

Solve sequentially:

Uvel Vvel Wvel

Solve pressure–correction

(continuity) equation

pressure, and velocity

turbulence, and other

scalar equations

No Yes

Converged? Stop

Fluent [9].

an algorithm to solve for the governing equations in a sequential order with an

iterative process as shown in figure 3.

3.1.1.2 Discretization technique ANSYS Fluent utilizes a discretization

technique to turn a general scalar equation into an algebraic equation which

enables the equations to be solved numerically. The governing equations are

integrated about each of the volumes created during the meshing process.

Nfaces

Nfaces

f

f v f f A f f A f SV

f

(6)

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 21

where

N faces number of faces enclosing the cell

f value of convected through the face f

f v f A f mass flux through the face f

A f area of face f

f gradient of at face f

V cell volume

requires ways to measure the validity and accuracy of the simulated solution.

The convergence criteria in ANSYS Fluent depend on the type of model chosen.

These residuals, depending on the type of model selected, involve x- and y-

components of the velocities, k and ε that include continuity, momentum,

turbulence, and energy.

When modeling the fluid domain surrounding a solar panel using the CFD

software, incompressible, steady, Newtonian fluid was assumed of a turbulent

nature with no buoyancy effects and no heat considerations. The layout of the

flow fields for two computer models that include five solar panels and five solar

panels with a deflector at the front are shown in figures 4(a) and 4(b). The

reduction in the wind loads on the panel using the wind deflector as predicted by

CFD results are discussed in a subsequent section below.

(a)

Wind

flow

(b)

Wind

flow

Figure 4: Layout of the flow field for CFD simulation: (a) five solar panels,

and (b) five solar panels with a deflector at the front [8].

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

22 Advances in Fluid Mechanics IX

It is a well-known fact that CFD analysis alone is not an absolute measure of the

accuracy in the design of any turbulent flow management system due to inherent

uncertainty in the turbulence parameters in the flow field. However, comparisons

between results from wind tunnel tests and CFD simulations for a few

benchmark models provide the degree of confidence required in the

implementation of the design at reduced cost and time. As full scale models were

too large to fit into our wind tunnel test section, quarter-scale models were

developed and tested in our university’s wind tunnel facility. All full-scale solar

panel racks that were investigated in this study were at an inclination of 10o with

respect to the horizontal axis for a wind speed of 49 m/s (110 mph).

The experimental analysis consisted of conducting multiple wind tunnel tests of

quarter-scale rooftop solar panel racks and wind deflectors. Figure 5 shows the

schematic of the test setup. The fan motor speed was controlled using VFD

(PowerFlex-4M, Allen-Bradley) to generate variable input frequencies to the

motor that translates into variable wind speeds in the tunnel as inlet wind speed

The VFD was actuated via a laptop (Dell E5500) utilizing LABVIEW v8.6

drivers and National Instruments’ module NI 9264 by providing voltage in the

range of 0~10 Vdc. At each motor rpm, the pitot- static tube that was connected

to an electronic pressure sensor (Model 20 INCH D-MV R8B22-58 supplied by

All Sensors) measured the pressure difference (P0 – P∞), where P0 is the

stagnation pressure and P∞ is the static pressure in the wind tunnel. The pressure

sensor outputs a dc voltage to NI 9219 with a supplied excitation voltage of 2.5

Vdc. Then through a calibration procedure provided by All Sensors, the voltage

was converted to a pressure difference in inches of water.

Through Bernoulli’s equation, the pressure difference was converted to wind

speed in mph. The NI modules interfaced with the NI DAQ module cDAQ-9172

and were controlled by the LABVIEW program. The wind tunnel was calibrated

by recording wind speeds at various input frequencies from 0 to 80 Hz in order

to interpolate between the wind speed and the corresponding input frequency.

The models to be tested were to be mounted on a melamine board. Four load

cells (FUTEK’s model LCF300:50-lb) were used to record either lift or drag

forces. The load cells operate on a full Wheatstone bridge. NI 9237 provided the

bridge circuitry. Through a calibration equation provided by FUTEK, the output

dc voltage was converted to a force in lbf and then to N through LABVIEW.

The load cells were installed with two orientations in order to measure the two

major wind forces; namely lift (vertical) and drag (horizontal) forces, for several

incoming wind speeds. Following figure 5, figures 6 to 8 below show the wind

tunnel and solar panel/load cell setup for wind lift and drag force measurements.

The results that were obtained from the wind tunnel analysis were then used to

verify the CFD simulations.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 23

Deflector

Wind

flow

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

24 Advances in Fluid Mechanics IX

(a) (b)

Figure 7: Load cell placement: (a) horizontal for the lift; (b) vertical for the

drag force measurement.

Figure 8: Wind tunnel instrumentation showing load cells on the testing floor

(zoomed and shown) for wind uplift force measurement.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 25

The results we report here are presented in the two subsections below. First the

wind loads measured for scaled models in the wind tunnel and predicted by CFD

analysis are presented followed by the CFD results for the full scale model only.

Measurements were recorded at four different speeds: 6.7, 9.0, 11.2, and 12.0

m/s (15, 20, 25, and 27 mph). For all the test configurations of the scaled models,

the lift and drag forces increased as the wind speed increased from 6.7 m/s (15

mph) to 12 m/s (27 mph) as expected due to the forces being proportional to the

square of wind speed. For rack-only arrangement, the minimum and maximum

lift forces were 5.34 N and 16.90 N respectively while for the rack-with-deflector

system the minimum and maximum lift forces were 2.67 N and 7.56 N

respectively. Therefore, it can be concluded solely from the wind tunnel test data

that using the deflector reduces the lift forces for the rack-deflector arrangement

by approximately 50% for the speeds tested while the drag forces have minimal

changes.

4.2 CFD simulation results compared with the wind tunnel test data

As shown in table 1, the use of the deflector reduced wind uplift by 51.29%, but

increased the drag force by 28.33%. It should be noted that positive values for

the forces represent a reduction and vice versa. The CFD analysis estimated a net

reduction of 37.14% and 9.16% for the lift and drag forces, respectively. A

similar trend was found in the experimental and computational analysis of the

three-rack model with and without deflector.

Table 1: Wind tunnel test data and CFD results for quarter-scale models

with input wind velocity=12 m/s – comparison in %.

For a single-rack model, the percentage reduction was 54.29% and 37.14%

for experimental and computational results, respectively. The percentage

reduction was even closer when the two methods were compared for the three-

rack model with deflector: 45.61% for experimental uplift and 54.09% for

computational uplift. In terms of drag results, the experimental results resulted in

slight additions to the drag force, for both single-rack and three-rack models

using the deflector, the addition was around 25%. This addition in wind drag is

expected as the addition of the deflector with the rack arrays will somewhat

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

26 Advances in Fluid Mechanics IX

increase the amount of drag force. When the respective drag and uplift forces are

compared for both the scenarios of single-rack only and three-rack array, drag

forces were ~70% smaller than the wind uplifts. This comparison explains that

the reduction in wind uplifts is more crucial than the wind drag forces for design

and installation purposes.

5 Conclusion

We conclude that an elliptic deflector with equally spaced short fins when placed

in front of the panel rack greatly reduces uplift on the panels. Wind tunnel

experiments showed a slight increase in drag force when a deflector was placed

at the front. This is due to the fact that the deflector acts as a bluff body

obstruction to the wind flow over the panel. There is a close agreement between

computational results and the wind tunnel test data in terms of the calculation of

lift forces; however, drag forces were not compared. Lift forces were mainly

responsible for the structural stability of solar panels under high winds. It was

also noted that a selective grid scheme and refinement in the flow domain have a

significant influence and improve CFD results, but lead to much longer

simulation run times and convergence. It was concluded that an elliptically

profiled wind deflector with uniformly spaced short fins, positioned before the

tilted panels, minimizes the high wind loads by as much as 50% compared to the

wind loads without its use. Details of the study are currently being reviewed and

will also be published in a journal paper.

Acknowledgement

This work was funded in part by Northern States Metal, Inc., Austintown, Ohio,

USA.

References

[1] Spratley, W.A., 1998, Solar Rooftops as Distributed Resources, The

Electricity Journal 11, pp. 40.

[2] Addressing http://www.nafcointernational.com on 02/01/2012 NAFCO

International Inc, 2006-2011, Fond Du Lac, WI, USA.

[3] Barkaszi, S.F. and Dunlop, J.P., 2001, Discussion of Strategies for Mounting

Photovoltaic Arrays on Rooftops, Proceedings of Solar Forum 2001, Solar

Energy: The Power to Choose, April 21-26, Washington D.C., pp. 6.

[4] Chung, K., Chang, K., Liu Y., 2008, Reduction of wind uplift of a solar

collector model Journal of Wind Engineering and Industrial Aerodynamics,

96, pp. 1294.

[5] Yatsco, M. 2011, Numerical Analysis and Wind Tunnel Validation of

Wind Deflectors for Rooftop Solar Panel Racks. Master's Thesis, School of

Graduate Studies & Research, Youngstown State University, Youngstown,

Ohio, USA.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 27

[6] Meroney, R.N., Neff, D.E., 2010, Wind effects on roof-mounted solar

photovoltaic arrays: CFD and wind-tunnel evaluation, The Fifth

Computational Wind Engineering (CWE2010), Chapel Hill, North Carolina,

May 23-27.

[7] Addressing http://www.extrusions.com/ on 02/01/2012 Northern States

Metals, Inc., Youngstown, OH, USA.

[8] Jiyuan Tu, Guan Heng Yeoh, and Chaoqun Liu, 2008, Computational Fluid

Dynamics: A Practical Approach, Butterworth-Heinemann.

[9] Addressing http://www.ansys.com/ on 02/01/2012 ANSYS Customer Portal

for Fluent Documentation and User Manuals.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

This page intentionally left blank

www.engbookspdf.com

Advances in Fluid Mechanics IX 29

a hemisphere

A. Goharzadeh, L. Khezzar & A. Molki

Department of Mechanical Engineering, The Petroleum Institute,

Abu Dhabi, UAE

Abstract

This paper presents an experimental study of laminar flow past a smoothed

hemisphere using the Particle Image Velocimetry technique. The experimental

setup consists of a rectangular channel, through which water flows over a 30-mm

diameter hemisphere mounted on a horizontal surface. From the measured

velocity distribution around the hemisphere a physical insight into the flow is

presented. Vertical and horizontal 2D velocity distributions are obtained for a

constant Reynolds number of Re=800, corresponding to a regime of laminar

flow. Measurements revealed the three dimensional structure of the flow

including a horseshoe vortex surrounding the hemisphere and arch-shaped

vortices in the downstream region. Both instantaneous and average velocity

distribution were studied. The location of the reattachment point, the separation

line, and the reverse flow are identified and discussed.

Keywords: PIV, horseshoe vortex, hemisphere, dome.

1 Introduction

Hemispherical shapes are widely used in many industrial applications such as

domed roofs in civil engineering to cover buildings or above-ground steel tanks

to store fluids. Domes were used in hydraulic channels as a mechanism to

generate and study the shedding of hairpin vortices in their wake [1]. Predicting

the three dimensional structure of flow field around a hemisphere is challenging

for engineers due to the complexity added by the curved surface. Therefore, in

order to ensure proper design of such engineering structures, detailed knowledge

of the flow structure around them is necessary.

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120031

www.engbookspdf.com

30 Advances in Fluid Mechanics IX

motions with separation and multiple re-attachment points, which at times

exhibit a highly unsteady behaviour. Thus, a small recirculating bubble exists

around the front stagnation point and a horseshoe vortex starts to develop and

envelopes parts of the dome. The wake is characterized by vortex shedding, flow

separation, and reattachment. Several parameters affect the flow behaviour

around domes: the dome shape, the Reynolds number (based on the approaching

free stream velocity and dome diameter), the inflow conditions such as the

approaching boundary layer shape, and the upstream-wall and dome surface

conditions.

Most of the experimental studies were conducted under turbulent flow

conditions. For example, Taniguchi et al. [2], Savory and Toy [3] and Cheng and

Fu [4] conducted experimental investigations to study the effect of three different

approaching boundary layers in addition to the effect of dome surface roughness

on the mean pressure distribution and on the critical Reynolds number beyond

which the pressure distribution becomes invariable. Tamai et al. [5] explored the

formation and shedding of vortices from a dome in a water tunnel and

determined the frequencies characterizing each phenomenon. As the maximum

Reynolds number reached 104, the Strouhal number had a value of 0.2.

Approaching the front side, successive expanding recirculating zones and the

formation of the well-known horseshoe vortices were observed. The number of

these structures increases with increasing Reynolds number until a critical value

of about 3,000, beyond which only one single recirculation zone persists with a

constant size. Behind the dome, arched vortex tubes form and are shed

individually or coalesce before shedding depending on a critical Reynolds

number of 2,000. Using split-fiber probes and hot-wire anemometers

Tavakol et al. [6] investigated the turbulent flow around a dome by obtaining

axial velocity profiles and its root mean square (RMS) component. In their

numerical investigations of laminar flow around a hemisphere, Kim and Choi [7]

found that the separation angle is fixed at 90° from the stagnation point at

Reynolds numbers below 300 and that the flow loses its symmetry and

steadiness with increasing Reynolds number with clear hairpin type vortex

shedding. The onset of asymmetry and periodicity occur at a Reynolds number

of between 170-180 and 190-200 respectively.

It appears that relatively few experimental studies were performed on laminar

flow around hemispherical domes; in other words, the majority being confined to

turbulent flows and where only point measurements were obtained. From these

measurements, extrapolated but incomplete pictures of the flow structure were

proposed. The flow around a dome is complex and three-dimensional and only

planar velocity measurements can provide a more complete picture of the flow

structure. Hence, the present work aims to fill this gap by reporting on velocity

measurements and flow mapping obtained by using Particle Image Velocimetry

(PIV) technique. The results are of sufficient accuracy to provide a benchmark

for the testing of computational methods of such flows [7].

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 31

2.1 Experimental setup

with the flow loop in a vertical configuration. The water tunnel consists of a

reservoir, head tower, flow conditioning section, contraction section, square

flume, and a downstream tower. The reservoir is of monolithic type, constructed

from clear acrylic, and contains the working fluid. Flow is generated using a

submersible pump (Grundfos Model KP 250). Flow conditioning is achieved by

means of a perforated cylinder which distributes the flow to the head tower, a

stainless steel perforated plate, and a series of acrylic screens. The contraction

section has a 2.52:1 area ratio with a symmetrical cross section and analytically

developed contours. The contraction guides the flow from the head tower to the

square flume which acts as the experimental test section.

As shown in figure 1a, a smooth hemispherical dome with a diameter of 30

mm was mounted in the test section and placed 220 mm downstream of the

contraction section. The dome was fabricated from acrylic using a conventional

lathe machine and painted in black colour to minimize reflection. Flow velocity

in the test section is estimated to vary from 15 to 300 mm/s with an uncertainty

of +/-2% and is controlled using a ball valve.

with a large field of view 90x70 mm2

In the central region of the channel, the flow over the hemisphere is laminar

and characterized by a constant fluid-based Reynolds number:

ܴ ൌ (1)

ఔ

where U=22 mm/s is the bulk stream velocity in the horizontal direction, D=30

mm the dome diameter and =0.8 mm2/s the kinematic viscosity of water at a

constant temperature of T=30°C corresponding to the Reynolds number of

Re=800.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

32 Advances in Fluid Mechanics IX

Using PIV, all flow field measurements were taken in two directions: (i) in the

central vertical axial plane of the dome arrangement, fig. 1a, and (ii) in the

horizontal plane near the bottom wall. To illuminate theses measurement planes,

a laser sheet was projected vertically and horizontally, with a charge coupled

device CCD camera positioned perpendicular to the plane to record particle

motion, as illustrated in figure 1a. The water was seeded with polyamide

particles tracers with an average diameter of 22 m and density of 1.016 g/cm3

for flow visualization. The PIV flow field measurements were undertaken at two

different locations: (i) upstream region, far from the dome location and (ii)

around the dome region. An example of a PIV image is shown in figure 1b,

which demonstrates the flow around the hemispheric obstacle. The flow was

observed with a field of view of 9070 mm2. The tracers’ density was 15–20

particles per interrogation area. A diode laser was used as the light source for

PIV measurements having a maximum power output of 200 W at a wavelength

of 532 nm. A cylindrical lens (f=28 mm) was used to generate a light sheet of

approximately 1 mm thickness.

The synchronization between laser light pulses and the camera was

accomplished by transistor-transistor logic pulses from the synchronizer. The

time interval between each pair of images was 200 ms with the pulse separation

time adjusted to 10 ms. Full-frame images of 10241024 pixels (figure 2b) were

acquired and transferred to a computer via a frame grabber. Using the DAVIS

FLOWMASTER software provided by the Lavision System (Germany), the 2D

PIV image was divided into 3232 pixels size sub-regions using a multi-grid

correlation process with 50% overlap. The average particle velocities were

calculated using 100 images coupled with the cross correlation method. The

spurious vectors calculated based on the local median filtering were less than

3%. No data smoothing was used for the calculation of the velocity

measurements.

60

40

y (mm)

20

0

0.015 m/s

-20 0 20 40 60

x (mm)

(a) Mean axial velocity at inlet. (b) Averaged axial velocity profiles

around the dome.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 33

To verify the upstream flow conditions, a cross stream averaged velocity profile

was obtained and is illustrated in figure 2a. The velocity profile reached a

maximum velocity between 0.2<y/ymax<0.8. The flow does represent a

developing type. The boundary layers are visible at the top and bottom side of

the channel.

In figure 2b, the large view of the averaged velocity profile around the dome

is presented for the Re=800. It is important to note that the flow is focused near

the bottom wall of the channel and therefore the velocity does not reach its

minimum value near top wall (0<y<60 mm).

The magnitude of particle velocities is constant far from the hemisphere with

an average velocity of 22 mm/s. Naturally and through the action of the pressure

field as the fluid particles approach the hemisphere, their velocities decrease.

20

y (mm)

10

-20 0 20 40 60

x (mm) 0.015 m/s

(a) Averaged axial velocity vector plot and its corresponding streamlines

in a mid vertical plane for Re=800.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

34 Advances in Fluid Mechanics IX

Streamlines of the averaged velocity field, are presented in figure 3a, and

compared with previous experimental observations reported by Savory and Toy

[3], fig. 3b. In this paper the flow is laminar (Re=800). The averaged velocity

profile shows parallel streamlines in the upstream region far from the dome

obstacle (figure 3a). As the flow approaches the dome the streamlines are

deviated in different directions. In lower layers behind the obstacle, the direction

of the velocity changes. An upstream recirculating region similar to the previous

observation [3] is denoted with a front stagnation streamline. Both the

reattachment zone and the secondary circulation are clearly visible using the

streamlines. The separation zone is located between (x=21 mm and y=12 mm).

Although similarities are observed between previous experimental results [3] and

the present PIV measurements, fundamental differences exist such as the size

and position of the recirculating zone. The PIV measurements indicate that the

center of the recirculating zone is located both further downstream of the dome

and at higher vertical position. This difference is due to the characteristics of the

fluid flow (laminar and confined flow in this experiment versus turbulent and

open flow in the previous study [3]). The other similarity which can be observed,

is at the front side of the dome obstacle. Horseshoe vortices are shown in both

cases, figs. 3a and b. However, in the present experiment the dimensions of the

horseshoe vortices are larger than in the previous study of [3] and represent

almost half of the dome diameter.

The magnitude of the vertical (Vy) and horizontal (Vx) velocities are

illustrated in figures 4a and 4b respectively. As shown in figure 4a, the

maximum axial velocity appears at x=10 mm and y=15 mm, where the velocity

is deviated from its original direction due to the presence of the dome as an

obstacle. The negative velocities represented in dark blue color (or dark gray

color in black and white print-B/W) correspond to the existence of horseshoe

vortices. The negative and positive velocities are characterized by the dark blue

(or dark gray in B/W print) and red colors (or medium gray in B/W print)

respectively. The horizontal light blue region represents the thin boundary zone

where the velocity is zero corresponding to the region where the main and

recirculating flows are separated.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 35

near the bottom wall. In this case, since the laser is projected horizontally only

part of the flow is optically accessible for measurements due to shadowing. The

laser illuminates the horizontal flow field at 0.1mm above the bottom wall as a

camera records PIV pictures from the top view. Measurements are compared

with the flow characteristics obtained by Savory and Toy [3] in which horseshoe

vortices are observed around the dome. The reattachment is also clearly shown

behind the dome. In particular, the bottom parts of the arch type downstream

vortices mentioned above are captured by the PIV measurements. However, the

flow field between two figures is slightly different. The horseshoe zone in front

and around the dome is larger in the present experiment.

horizontal plane for Re=800. Re=1.4105 [3].

bottom wall.

The PIV measurements show that the flow is highly three dimensional around

the dome. It was also observed that flow inside the horseshoe vortices and

reattachment zone is not stable. The location of the separation zones fluctuate

with time between two instantaneous images. However, the averaged velocity

shows a structure and a quasi-symmetric flow in the horizontal plane.

4 Conclusions

Laminar flow past a smooth hemisphere was investigated using the Particle

Image Velocimetry technique. Vertical and horizontal 2D velocity distributions

at Re=800, showed a laminar flow. The averaged velocity distributions around

the hemispherical obstacle were obtained and velocity field illustrated the

presence of a complex three dimensional flow containing intriguing structures

such as a horseshoe vortex enveloping the dome and arch-shaped vortices in the

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

36 Advances in Fluid Mechanics IX

downstream region. The location of the reattachment point, the separation line,

and the reverse flow are also identified and discussed. The characteristics of

laminar flow around the dome share several similar features with the case of

turbulent flow [3] with significant difference in the size of vortices and their

location. The results of this research can be employed to validate further

numerical predictions of this complex flow field. Future experiments will focus

on the influence of the dome surface roughness and velocity distribution of

turbulent flows around the hemispherical obstacle.

References

[1] Acarlar, M.S. and Smith C.R., A study of hairpin vortices in a laminar

boundary layer, Part1. Hairpin vortices generated by a hemisphere

protuberance. J. Fluid Mech., 175, pp. 1–41, 1987.

[2] Taniguchi, S., Sakamoto, H., Kiya, M. and Arie, M., Time-averaged

aerodynamic forces acting on a hemisphere immersed in a turbulent

boundary. J. Wind Eng. Ind. Aerod., 9, pp. 257–273, 1982.

[3] Savory, E. and Toy, N., Hemispheres and hemisphere-cylinders in turbulent

boundary layers. J. Wind Eng. Ind. Aerod., 23, pp. 345–364, 1986.

[4] Cheng, C.M. and Fu, C.L., Characteristic of wind loads on a hemispherical

dome in smooth flow and turbulent boundary layer flow. J. Wind Eng. Ind.

Aerod., 98, pp. 328–344, 2010.

[5] Tamai, N., Asaeda, T. and Tanaka, N., Vortex structures around a

hemispheric hump. Bound-Lay. Meteorol., 39, pp. 301–314, 1987.

[6] Tavakol, M.M., Yaghoubi, M. and Masoudi Motlagh, M.M., Air flow

aerodynamic on a wall mounted hemisphere for various turbulent boundary

layers. Exp. Thermal and Fluid Sci., 34, pp. 538–553, 2010.

[7] Kim, D. and Choi, H., Laminar flow past a hemisphere. Phy. of Fluids,

15(8), pp. 2457-2460, 2003.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 37

a CFD analysis

F. Concli & C. Gorla

Politecnico di Milano, Department of Mechanical Engineering, Milano,

Italy

Abstract

Efficiency is becoming more and more of a main concern in the design of power

transmissions and the demand for high efficiency gearboxes is continuously

increasing; also the euro standards for the reduction of pollutant emissions from

light vehicles imposed to improve the efficiency of the engines and gear

transmissions are becoming more and more restrictive. For this reason the

resources dedicated to this goal are continuously increasing.

The first step to improve efficiency is to have appropriate models to compare

different design solutions. Even if the efficiency of transmissions is quite high

compared to the efficiency of the engines and appropriate models to predict the

power losses due to gear meshing, to bearings and to seals that already exist, in

order to have further improvement, some aspects like the power losses related to

the oil churning, oil squeezing and windage are still to be investigated. In

previous papers, the authors have investigated by means of CFD (computational

fluid dynamic) analysis and experimental measurements the churning losses of

planetary speed reducers (in which there is a relative motion between the

“planets + planet carrier” and the lubricant). This report is focused on the oil

squeezing power loss. This kind of loss is associated with the compression-

expansion process by the meshing teeth. The contraction of the volume at the

gear mesh implies an overpressure that induces a fluid flow primarily in the axial

direction and this, for viscous fluids, means additional power losses and a

decrease of efficiency.

In this work this phenomenon has been studied by means of some CFD

simulations. The influence of some operating conditions, such as the lubricant

properties, rotational speed and temperature, have been studied.

Keywords: efficiency, gear, power losses, oil squeezing, lubrication, CFD.

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120041

www.engbookspdf.com

38 Advances in Fluid Mechanics IX

1 Introduction

As the fuel economy requirements and euro standards for the reduction of

pollutant emissions are becoming more and more stringent, efficiency is

becoming a main concern also in the design of power transmissions. Reducing

fuel consumption and particulate emission levels through increasing efficiency

has not only a strong impact on the economy and on the environment, but also on

the reliability on the transmissions: a reduction of the power losses means also a

reduction of the heat generated and, therefore, of the operating temperature of the

transmission. A lower temperature is favorable for the system reliability.

Sources of losses in a gearbox can be classified according to [1] into two

main categories: (i) load dependent power losses and (ii) load independent power

losses. The load dependent power losses are primarily related to a mechanical

power loss due to friction at the gear contact and between the rolling elements

and the races of the bearings. The load independent power losses, in turn, are

primarily related to viscous effects. These losses can be further subdivided into

oil churning and windage losses that are the result of the interaction between the

oil/air and the moving/rotating elements like gears and shafts, into

pocketing/squeezing losses due to the pumping effect of the mating gears and

into other viscous dissipations like those of the bearings.

In order to predict these kinds of losses, literature provides many publications

about the load dependent power losses, but only few works about load

independent power losses. These works are primarily concerning the churning

losses of ordinary gears, the spin power losses and the losses of the bearings.

What is still wanted are appropriate studies to predict the oil pumping/squeezing

power losses. The works on this topic are few and simplified. Strasser [10]

proposed a simplified fluid-dynamic model to predict the oil squeezing power

losses in a spur gear pair. This model takes into account the variation of the gap

between the teeth but it approximates the geometry. Seetharaman and

Kahraman [11] proposed a physic based fluid mechanics model to predict the

spin power losses of a gear pair due to oil churning and windage. This model

calculate the losses like the sum of the losses associated to the interactions of

individual gears and the fluid and the power losses due to pumping of the oil at

the gear mesh. To predict the oil pumping losses, Seetharaman and Kahraman

developed an analytical model that take into account the variation of the volume

of the gap between the teeth without simplifying the geometry like in the model

proposed by Strasser. These models are able to give results in a very short time

but they give only approximate results. For this reason the authors have

investigated this kind of losses by mean of CFD simulations that are able to

correctly solve the velocity and the pressure fields without simplifications and,

therefore, give more information about the phenomena. In this study the

influence on this kind of losses of the oil temperature and of the rotational speed

of the gears have been investigated.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 39

2 Problem description

Between two mating gears, the volume of the cavity between the teeth is

continuously changing. The sudden contraction of this volume caused by the

gears revolution implies an overpressure in the gap. The lubricant is therefore

squeezed out primary in the axial directions. After the reaching of the minimum

value for the volume, this increase again causing a negative pressure in the gap

and, consequently, a fluid flow takes place from the oil bath to the cavity

between the teeth.

This process is cyclic and, in general, there are multiple cavities that are

squeezed together. Due to the viscous properties of the lubricant, this

phenomenon induces power losses.

Figure 1 illustrates the phenomena: the volume (marked in grey) decreases

from picture 1a (where a gear pair starts the contact) to picture 1b. Figure 1c

shows the moment in which a second gear pair mates and figure 1d the quick

growth of the cavity volume.

Figure 1: Change of the volume of the cavity during the engaging of the

gears.

3 Geometry

In order to study this kind of power losses, a spur gear pair has been used. Pinion

and gear have the same dimensions. Table 1 summarized the main parameter of

the gears.

For this initial study, a full immersion lubrication has been adopted. This

solution has been chosen in order to simplify the model. A partial immersion, in

fact, complicates a lot the problem because it involves multiphase flows. The

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

40 Advances in Fluid Mechanics IX

aim of this initial study was to understand the effect of the mating gears on the

axial flows, so the full immersion lubrication was considered appropriate at this

point of the research.

helix angle on

Pitch circle r

reference 0 14,4

[mm]

cylinderb[rad]

Working pitch

Pressure angle ' [rad] 0,358452 14,45

circle r' [mm]

Working pressure

0,349066 Number of teeth z 36

angle [rad]

Centre distance a Base circle rb

28,8 13,5315

[mm] [mm]

Working centre

28,9 Module m [mm] 0,8

distance a' [mm]

Face width b

Tip radius ra [mm] 15,175 15,5

[mm]

Root circle ri [mm] 13,4375

4 Geometrical model

In order to reduce the number of cells and thank to the symmetry of the gear pair,

only half domain has been modeled. The aim of this work was to study in detail

the behavior of the lubricant in the region of meshing. For this reason only five

teeth pro gear have been modeled.

The computational domain for the CFD analysis has been modeled by means

of 3D cad software and discretized with a swept mesh. This meshing technique

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 41

consists in creating a mesh on one side of the region, known as the source side,

and then copying the nodes of that mesh, one element layer at a time, until the

end side, known as the target side, is reached.

The whole model has been discretized with triangular prisms. This kind of

elements allows a larger aspect ratio compared with the tetrahedral cells in which

it will invariably affects the skewness of the cell, which is undesirable as it may

affect accuracy and convergence. Some mesh refinements have been performed

in order to find the best compromise between the accuracy of the solution and

the computational time. The real problem during the meshing operation is that

it’s necessary to model the thin oil film that rise between the mating teeth. The

thickness of this film has a big influence on the results in terms of power losses.

For this reason some geometries with different film thickness have been

modelled.

After a convergence study, the final model had a film thickness of about 0.5

m. That means that the required mesh is extra fine. The smallest element has a

length of about 0.2 m.

maximum volume [m³] 5.51E-03

total volume [m³] 3.14E+01

minimum face area [m²] 1.60E-06

maximum face area [m²] 6.03E+00

minimum orthogonal quality 0.38633

5 Numerical model

To simulate the problem, a VOF approach has been used. This method is an

Eulerian method characterized by a mesh that is (in this case) moving in a certain

prescribed manner to accommodate the evolving shape of the interface.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

42 Advances in Fluid Mechanics IX

The governing equations are the conservation equations for mass and

momentum. The energy equation is not activated in these simulations.

∙ (1)

∙ ∙ (2)

is the density, is the time, is the velocity vector, is the pressure, and

are the gravitational body force and external body forces.

For compressible flows, the unknowns are the velocity components and the

density while the pressure is evaluated with a constitutive equation. For

incompressible flows, the variables are the pressure and the velocity

components. The solution of the system of equations for incompressible flows is

accomplished by the fact that there are not equations where the pressure is

explicity defined. To calcluate it, the continuity equation is substituted with an

equation for the pressure; with some manipulation the pressure appears like

unknown term in the momentum equation.

For this reasons a SIMPLE (Semi Implicit Method for Pressure-Linked

Equations) scheme has been adopted as suggested for flows in closed domains to

solve the pressure-velocity-couplig. This algorithm uses a relationship between

velocity and pressure corrections to enforce mass conservation and to obtain the

pressure field. The idea is that the pressure field must guarantee, at every time,

the continuity equations. Is it therefore necessary to obtain a diefferntial equation

for the pressure unknown derived from continuity and momentum equations.

In this method the pressure field is calculated as

∗

′ (3)

∗

where is the estimated pressure guessed or obtained at the previous time

step/iteration from the momentum equations and ′ is the pressure correction

The velocity field is calculated as

∗

(4)

where ∗ is the estimated velocity calculated from the pressure ∗ and ′ is the

velocity correction.

The gear surfaces have been set like no slip walls and the lateral sides of the

domain as zero pressure to simulate the steady oil bath and as symmetry

respectively.

In order to reproduce the operating conditions, a rigid motion of the

boundaries corresponding to the pinion and the gear has been applied by means

of User Defined Functions (UDF) written in the c++ language. With this

boundary motion it is possible to reproduce the real operating conditions in

which gears are mating together. The time step for the transient analysis has been

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 43

chosen as dynamic. This means that every step, the new time increment depends

on the velocity field.

,

(5)

and is the velocity scale of the problem. This allows us to increase the

calculations without the risk to lose the convergence.

This boundary motion implies a deformation of the fluid domain. For this

reason it is necessary do update the mesh every time step. In order to do that, a

dynamic mesh model has been adopted: the spring-based smoothing +

remeshing. In this method, the edges between any two mesh nodes are idealized

as a network of interconnected springs. The initial spacing of the edges before

any boundary motion constitutes the equilibrium state of the mesh. A

displacement at a given boundary node will generate a force proportional to the

displacement along all the springs connected to the node. With a linear elastic

behavior, the force on a mesh node can be written as

∑ ∆ ∆ (6)

where ∆ and ∆ are the displacements of node and its neighbor , is the

number of the neighboring nodes connected to the node and is the spring

constant between node and its neighbor . The spring constant for the edge

connecting nodes and is defined as

(7)

| |

At equilibrium, the net force on a node due to all the springs connected to the

node must be zero. This condition results in an iterative equation such that

∑ ∆

∆ (8)

∑

using a Jacobi sweep on all interior nodes. At convergence, the positions are

updated such that

,

∆ (9)

where 1 and re used to denote the positions at the next time step and the

current time step, respectively.

Figure 4 shows the adopted mesh before and after the calculations. It is

possible to appreciate the good efficiency of the smoothing-based algorithm for

the mesh update. After the iterations, the elements have still a good aspect ratio

and approximately the same dimensions that they had at the beginning.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

44 Advances in Fluid Mechanics IX

a) b)

Figure 4: a) Detail of the mesh at the first time step; b) Detail of the mesh

after some iterations.

6 Operating conditions

The purpose of the simulations is to calculate the power losses due to oil

squeezing under different operating conditions.

In order to do that, the resistant torque on the driving shaft has been

monitored. This resistant torque is calculated with a surface integral on the

moving walls with respect to the gearbox axis and it is composed of two parts:

the first given by the pressure and the second by the viscous effects. Starting

from the resistant torque it is possible to calculate the power losses just by

multiplying it by the rotational speed. The simulations have been computed with

different combinations of operating temperature (which means different density

and viscosity) and rotational speed.

Table 3 shows the combinations of parameters for each simulation. is the

operating temperature in °C, the rotational speed of the planet carrier in

[rad/s], the density in [Kg/m³] and the dynamic viscosity in [Pa*s].

Rotational speed Temperature Density Dynamic Viscosity

[rad/s] T [°C] [kg/m³] [Pa*s]

52 40 1040.86 0.2289

105 40 1040.86 0.2289

157 40 1040.86 0.2289

52 65 1021.61 0.1481

105 65 1021.61 0.1481

157 65 1021.61 0.1481

52 90 1002.36 0.0702

105 90 1002.36 0.0702

157 90 1002.36 0.0702

A commercial oil whit this characteristics can be, for example, the

Klübersynth GH 6-220 Synthetic Gear Oil.

7 Results

Figure 5a shows the pressure field on the symmetry plane. It is possible to

appreciate that in the upper cavity the pressure is higher than in the oil bath.

That’s because this cavity is decreasing his volume.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 45

The lower cavity, in turn, shows a lower pressure depending on the fact that it

is increasing his volume.

Figure 5b shows the velocity distribution in the axial direction. It can be

appreciate the flux in the axial direction.

a) b)

distribution in the field of action.

Figure 6 to 8 show the results in terms of power losses on the driving wheel

for different operating conditions and for 1 cycle. Since gear mesh is, also the

resistant torque necessary to squeeze the oil has a periodic trend.

a) b)

c)

Figure 6: Power losses on the driving shaft; a) 40°C – 52rad/s;

b) 65°C – 52rad/s; c) 90°C – 52rad/s.

From the diagrams it can be seen that, as expected, the resistant torque

decreases with temperature while it increases with rotational speed. It can be also

seen that this kind of losses are low in the case of the highest temperature and the

lower speed (figure 6c), in which they are less than 1 W, while in the case of

lowest temperature and higher speed (figure 8a) they became significant.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

46 Advances in Fluid Mechanics IX

a) b)

c)

b) 65°C – 105rad/s; c) 90°C – 105rad/s.

a) b)

c)

b) 65°C – 157rad/s; c) 90°C – 157rad/s.

Figure 9 shows the power losses along the contact path. In the position 1 the

first teeth pair (1s+2s), as shown in figure 10a, is already engaged. The second

teeth pair (2s+3s) is starting the contact. In the position 2 of figure 8, the first

teeth pair ended the contact (figure 10b).

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 47

Between 1 and 2, the mating teeth pairs are therefore 2. In the position 3,

another teeth pair (3s+4d) starts the contact (figure 10c) while the second teeth

pair (2s+3s) is already engaged. Like in position 2, in position 4 a teeth pair

(2s+3s) ended the contact (figure 10d).

From this position to position 5, the engaged teeth pair is again only one, like

between position 2 and position 3. Finally, in position 5, like in position 3, a

teeth pair starts the contact while another teeth pair is already engaged (figure

10e).

8 Conclusions

As reliable models to predict the oil squeezing losses at the gear mating are still

not available, a CFD model has been applied in order to predict this important

component of losses. The simulations show that this kind of losses increase, as

expected, with the rotational speed and decrease with temperature. The results

show also that for low regimes and high temperature the kind of losses are

particularly low.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

48 Advances in Fluid Mechanics IX

Future works will be the study of other geometries and other kinds of

lubrications.

References

[1] Niemann, G., Winter, H., Maschinenelemente – Band 2: Getriebe

allgemein, Zahnradgetriebe – Grundlagen, Stirnradgetriebe –– 2.Auflage –

Springer, Berlin 2003.

[2] ISO/TR 14179-2:2001(E) – Gears – Thermal capacity – Part 2: Thermal

load-carrying capacity.

[3] Catalogo generale SKF – Edizione scolastica – Gruppo SKF, Dicembre

2006.

[4] Concli, F., Gorla, C., Arigoni, R., Cognigni, E., Musolesi, M.,:” Planetary

Speed Reducers: Efficiency, Backlash, Stiffness”, International conference

on gears, Munich 2010.

[5] Concli, F., Gorla, C., Arigoni, R., Musolesi, M.:” Riduttori di precisione a

gioco ridotto ed alta efficienza”, Organi di trasmissione – febbraio 2011,

Tecniche Nuove, Milano 2011.

[6] Concli, F., Gorla, C.:”Computational and experimental analysis of the

churning power losses in an industrial planetary speed reducers”,

Multiphase flow IV, Wessex Institute of technology, 2011.

[7] Csobàn, A., Kozma, M., Influence of the Oil Churning, the Bearing and the

Tooth Friction Losses on the Efficiency of Planetary Gears, Journal of

Mechanichal Engineering 56(2010)4, pp. 231-238.

[8] Concli,F, Gorla,C, Churning power losses in planetary speed reducer:

computational-experimental analysis, EngineSOFT International

Conference 2012 Conference Proceedings, 2011.

[9] Concli,F, Gorla,C, Influence of lubricant temperature, lubricant level and

rotational speed on the churning power losses in an industrial planetary

speed reducer: computational and experimental study - International

Journal of Computational Methods and Experimental Measurements –

Wssex Institute of Technology.

[10] D. Strasser, Einfluss des Zahnflanken- und Zahnkopfspieles auf die

Leerlaufverlustleistung von Zahnradgetrieben – Dissertation zur Erlangung

del Grades Doktor-Ingenieur – Bochum 2005.

[11] Seetharaman, A. Kahraman, Load-Independent Spin Power Losses of a

Spur Gear Pair: Model Formulation, Journal of Tribology APRIL 2009,

Vol. 131.

[12] Patankar, S.V., Numerical heat transfer and fluid flow, Taylor & Francis,

USA 1980.

[13] Versteeg, H.K., Malalasekera, W., An introduction to computational fluid

dynamics – The finite volume method, Longman Group, London 1995.

[14] Comini, G., Fondamenti di termofluidodinamica computazionale,

SGEditoriali, Padova 2004.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 49

at Derbendikan power station

R. A. Saeed1, V. Popov2 & A. N. Galybin2,3

1

University of Sulaimani, Iraq

2

Wessex Institute of Technology, UK

3

Institute of Physics of the Earth, Russia

Abstract

This paper presents the results of simulations of the complete three-dimensional

fluid flow through the Spiral Casing, Stay Vane, Guide Vane, and then through

the Francis turbine runner of the Derbendikan power station. To investigate the

flow in the Francis turbine and also to identify the loads acting on the turbine

blades, a three-dimensional model was prepared according to specifications

provided. The results show that the maximum absolute pressure of 4.4x105 Pa is

reached at the leading edge of the turbine runner and the maximum tangential

velocity reaches 33 m/s in the turbine runner.

Keywords: Francis turbine runner, numerical simulation, CFD modelling.

1 Introduction

Generally the flow field in Francis turbines is quite complicated due to its three-

dimensional nature and the curvature of the passages between runner blades. The

complexity of the problem requires the application of numerical approaches

which are capable of producing accurate results for flow velocity field and

pressure distributions on the runner. The application of CFD is an efficient way

for the analysis of fluid flow through hydraulic turbines [1]. Nava et al. [2]

presented an application of CFD to compute the loads caused by water pressure

on a blade of the Francis turbine runner. There is a satisfactory agreement

between numerical simulations and experimental measurements as reported by

Farhat et al. [4], who compared the flow simulation results and experimental

data for the whole Francis turbine. Three-dimensional steady flow analysis of the

flow passages from the entrance runner to the outlet in Francis turbine type has

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120051

www.engbookspdf.com

50 Advances in Fluid Mechanics IX

been reported by Ikeda et al. [5]. This analysis provides the distributed pressure

and water velocity within the computational domain.

In this paper, the pressure distribution and flow velocity field were

investigated by using Computational Fluid Dynamics (CFD) for specific

operating condition. Detailed analysis of the fluid flow through the turbine

runner is necessary in order to study the fluid flow conditions. For this purpose

the Spiral Casing, Stay Vane, Guide Vane, Runner and upper part of the Draft

tube have been considered. The corresponding model is shown in Figure 1.

This work includes the analysis of the performance of the Francis turbine of

Unit 2 in Derbendikan hydropower station; one of the major suppliers of

electrical power generation in the north of Iraq- Kurdistan Region. The

investigated turbine has the following data [7]: number of runner blades 13,

number of guide vanes 24, rated head 80 m, power output at rated head 83 MW,

discharge at rated head 113 m3/s, rotational speed 187.5 rpm.

2 Flow simulations

The flow simulation of the Francis turbine runner is quite complicated and can

be calculated only by using numerical methods. CFD simulation of the Francis

turbine runner was performed to obtain pressure distribution and flow velocity

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 51

through the turbine runner. Following previous research [9, 10], for the

computational domain, the Spiral Casing, Stay Vane, Guide Vane, Runner and

upper part of the Draft tube is considered.

The first step of flow simulation is building the geometrical model of the flow

domain. According to the provided specifications from Derbendikan hydropower

station a three-dimensional geometrical model has been created, as shown in

Figures 1 and 2. The geometry of the fluid domain has been created on

AutoCAD software and inserted into ANSYS software.

As shown in the figure, the entire fluid passageway between the inlet from the

Spiral Case side and the outlet from the draft tube side for the turbine is

considered.

The flow in the runner was computed in the rotating frame of reference, while

the flow in the stationary components was calculated in the stationary frame of

reference. Figure 2 shows the detail of the connection between flow domains

used for CFD simulations and Francis turbine runner model.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

52 Advances in Fluid Mechanics IX

In this study, three-dimensional discretization has been used with the finite

volume method (FVM) provided by the ANSYS CFX software. For the

computational domain, unstructured 3D tetrahedral meshing has been employed,

due to its flexibility when solving complex geometries, as shown in Figure 3.

The whole computational mesh consists of 3,241,986 volumes and 628,677

nodes.

paper focuses on 3-D Navier-Stokes simulations using the commercial code

ANSYS CFX. The continuity equation and Reynolds-averaged Navier-Stokes

(RANS) equation for incompressible flow have been used as governing equations

in the following form [11]:

U i (1)

0

X i

U i U i 1 P U U j (2)

U j i

ij 0

t X j X i X j X j X i

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 53

respectively, and ij are the components of the viscous stress tensor, also called

the Reynolds Stress Tensor. The turbulent effects on the flow field are taken into

account through the Reynolds Stresses ij , those are calculated from the k-ε

turbulence model, which is frequently used for modelling turbulent flow.

normal direction in the entrance of Spiral Case. Total mass flow in stationary

frame was specified at the inlet while static pressure was defined at the outlet

boundary. Inlet boundary conditions at given performance point of the hydraulic

turbine runner are derived from an operation condition which include discharge

Q (82 m3/s), where the Guide Vanes opening is 80%. Outlet boundary condition

is defined to an opening with an average relative pressure to atmospheric

pressure.

acceptable level, the residual is usually set between four to six orders of

magnitude lower than the actual values [12]. Iterative methods are used to solve

the corresponding system of algebraic equations, which allows one to

characterize the accuracy of the approximation solution by analysing the

residual. In this study, the RMS residuals settled to a steady state value within

100 iterations. The non-dimensional residual of the momentum and continuity

equations was between 10-4 to 10-6.

Three-dimensional analyses of fluid flow through the Spiral Case, Stay and

Guide Vanes and then through the turbine blades channel have been carried out.

The distributions of pressure and water velocity within the computational

domain are illustrated in the form of contour and streamlines that are generated

with the ANSYS CFX postprocessor.

The variation of pressure distributions and tangential velocity in the

computational domain obtained from the CFD analysis for specific operating

conditions are plotted in a plane which is passing through the centre of Spiral

Case and across the axis’s of rotation as shown in Figure 3(A and B).

The symmetrical inflow to the runner in circumferential direction can be

observed, which results in symmetrical distribution pressure from the uniform

flow distribution of the runner inlet, as shown in Figure 4B. The tangential

velocity increase has been detected at the lower ring, which is due to the smaller

area at the distributor region, as evident in Figure 4A.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

54 Advances in Fluid Mechanics IX

distributions [Pa] in the computational domain.

Figure 5(A and B), show the streamlines of the tangential velocity and the

runner surface pressure distributions in the plane passing through the centre of

Spiral Case and parallel to the axis’s of rotation. The tangential velocity

component decreases on the blade at the leading edge, which is due to the

sharp.bend of the flow in the stagnation regions near the blades, as observed in

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 55

Figure 5A. From Figure 5B, reduction from high pressure in the runner inlet to

low pressure at the runner outlet is clearly observed. High magnitudes of water

pressure are observed at the leading edge of the runner due to stagnation pressure

on the blades, which results from the rapid decrease in the magnitude of the

tangential velocity close to the bend of the flow.

Figure 5: (A) Tangential velocity [ms-1] and (B) the distribution of pressure

[Pa] in the turbine.

At the top of the runner the pressure is quite high whereas at the bottom the

pressure is low. The increasing pressure on the blades is clearly visible due to

stagnation. The reduction of pressure on the runner blade suction side is

explained by higher magnitudes of the velocity at the trailing edge. Also, the

pressure differences from the pressure and suction side are high and these

differences provide the torque on the shaft.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

56 Advances in Fluid Mechanics IX

As illustrated in Figure 6, the flow is coming out from the spiral inlet to the

upper part of the draft tube at the outlet. The flow is streamlined up to the runner

outlet and recirculation is established in the draft tube. This is due to the change

in the direction of the flow in the draft tube.

region of the runner and it is nearly axial at the centre. While the recirculation

region varies with the inlet boundary conditions, it increases gradually by

decreasing velocity at the inlet. There is a sharp increase in the magnitude of

velocity close to the trailing edge of the blade near the ring, due to the reduction

of the area between the blade and the ring.

3 Conclusions

The simulation of the 3D steady-state fluid flow with moving reference frame in

the entire Francis turbine from spiral case through stay vanes, wicket gates, and

runner down to the draft tube is presented. The flow field simulation needs to

include the entire flow passage to accurately model the runner surface pressure

distributions and the streamlines of tangential velocity. Both these characteristics

have been examined for a specific operation condition. Numerical simulation of

the entire Francis turbine flow shows clearly the influence of the operation

condition on the variation of pressure distributions and tangential velocity in the

computational domain. The results show good agreement with the previous

studies. The pressure profiles are used further to study stresses in the runner.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 57

References

[1] Avellan, F., Etter, S., Gummer, J. H., Seidel, U., 2000, “Dynamic

Pressure Measurements on a Model Turbine Runner and Their Use in

Preventing Runner Fatigue Failure”, 20th IAHR Symposium, Charlotte,

North Carolina, USA.

[2] Nava, J. M. F., Gómez, O. D., Hernández J. A. R. L., Flow induced

stresses in a Francis runner using ANSYS, International ANSYS

Conference Proceedings, 2006.

[3] Bjorndal, H., Moltubakk, T., Aunemo, H., 2001, “Flow Induced Stresses

in a Medium Head Francis Runner - Strain gauge measurements in an

operating plant and comparison with Finite Element Analysis”, 10th

International Meeting of the IAHR Work Group on the Behaviour of

Hydraulic Machinery under Steady Oscillatory Conditions, Trondheim,

Norway.

[4] Farhat, M., S. Natal, Avellan, F., Paquet, F., Lowys, Py., Couston, M.,

2002, “Onboard Measurements of Pressure and Strain Fluctuations in a

Model of Low Head Francis Turbine Part 2 : Measurements and

Preliminary Analysis Results”, Proceedings of the 21st IAHR Symposium

on Hydraulic Machinery and Systems, Lausanne, pp. 873-880.

[5] Ikeda, K., Inagaki, M., Niikura, K., Oshima, K., 700-m 400-MW Class

Ultrahigh-head Pump Turbine, Hitachi Review Vol. 49, No. 2, pp. 81-87,

2000

[6] Farhat, M., Avellan, F., Seidel, U., 2002, “Pressure Fluctuation

Measurements in Hydro Turbine Models”, 9th International Symposium

on Transport Phenomena and Dynamics of Rotating Machinery,

Honolulu, Hawaii, USA.

[7] Fattah, S. S., Khoshnaw, F. M., Saeed, R. A., 2005, “An Investigation

about Preventing Cavitation Damage and Fatigue Failure in

Derbendikhan Power Station”, Fluid Structure Interaction and Moving

Boundary Problems, Spain, Vol. 84, pp. 97-106.

[8] Friziger, J. H., 2002. Computational Methods for Fluid Dynamics.

Springer, New York, USA.

[9] Čarija Z., Mrša, Z., 2003, “Complete Francis Turbine Flow Simulation

for the Whole Range of Discharges”, 4th International Congress, of

Croatian Society of Mechanics, Bizovac, Croatia, pp. 105-111.

[10] Ruofu, X., Zhengwei, W., Yongyao, L., 2008, “Dynamic Stresses in a

Francis Turbine Runner Based on Fluid-Structure Interaction Analysis”,

Tsinghua Science and Technology, Vol. 13, No. 5, pp. 587-592.

[11] Ruprecht, A., Heitele, M., Helmrich, T., Moser, W., Aschenbrenner, T.,

2000, “Numerical simulation of a complete Francis turbine including

unsteady rotor/stator interactions”, Proceedings of the 20th IAHR.

[12] Hirsch, C., 2007, Numerical Computation of Internal and External Flows,

John Wiley and Sons, Ltd, Oxford, Great Britain.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

This page intentionally left blank

www.engbookspdf.com

Advances in Fluid Mechanics IX 59

a length to depth ratio of 4

Y. Y. Pey1, L. P. Chua1 & W. L. Siauw2

1

Department of Mechanical and Aerospace Engineering,

Nanyang Technological University, Singapore

2

DSO National Laboratories, Singapore

Abstract

The objective of this study is to characterize the flow behavior of an internal

cavity of L/D=4, with its shear layer separated from a thick boundary layer at a

free stream velocity of 15m/s. The study includes examining the cavity surface

pressure and the associated flow structures. Pressure measurements were

acquired from the cavity walls and the associated internal flow structures were

studied using Particle Image Velocimetry (PIV). Several data processing

techniques to study the flow structures were performed on the PIV data. These

include time averaging quantities: velocities, vorticity and spatial correlation. In

order to estimate the dynamics of the flow, the Proper Orthogonal

Decomposition (POD) was applied to 900 independent PIV acquired velocities

data set. The cavity wall surface pressure distribution indicated an open flow

structure and was consistent with the mean velocity measurements which

indicted a large re-circulating region within the cavity. Shear layer velocity

profiles across the cavity revealed self-similarity beyond a distance of x/L=0.3.

The shear layer growth rate of dθ/dx=0.031 was close to the entrainment rate of a

turbulent mixing layer, dθ/dx=0.035. Two point spatial correlation coefficients of

velocity fluctuations revealed a distinct pattern of alternating regions of positive

and negative coefficient values, indicating the presence of organized coherent

structures within the cavity. The modes one and two of the POD extracted

structures indicated a flow structure of the closed cavity type. Mode three

indicated the possibility of a large structure ejecting from the aft of the cavity.

Keywords: open cavity, coherent structures, proper orthogonal decomposition,

particle image velocimetry, spatial correlation, self-similarity, recirculation.

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120061

www.engbookspdf.com

60 Advances in Fluid Mechanics IX

1 Introduction

Study of flow over cavities is motivated by its phenomena found in many areas

such as in solar cells, slots between movable parts of ships, tandem arrangements

of bluff bodies such as adjacent tall buildings or tractor-trailer combinations, gas

dynamic lasers, hydraulic gates, control valves, landing gears of aircrafts,

sunroofs or windows in automobiles [1, 2]. Though geometrically simple, these

cavities are associated with complex phenomenon. When exposed to free stream,

the cavities experience unsteady flow around them. According to Rockwell et al

[3], aero acoustic tones are not present at low speed (Mach number less than

0.3); thus, unsteady flow behaviors are due to inherent interaction of the free

stream fluid with the lower speed fluid within the cavity. Cavities flow in general

can be defined as open, closed or transitional. In open cavity, the flow separates

at the leading edge of cavity and spans the entire length of the cavity. For a

closed cavity, the flow reattaches on the cavity floor downstream of the leading

edge before separating again near the downstream edge. A schematic diagram

depicting typical flow in open and closed cavity is shown in Figure 1. In the case

of transitional cavity flow, the separated shear layer attaches close to the aft of

the cavity.

Figure 1: Schematic diagram of open cavity (left) and closed cavity (right)

[4].

The current study was carried out on an open cavity with a length to depth ratio

(L/D) of 4 at free stream velocity of 15m/s. It includes the measurement of

pressure on cavity walls and the study of cavity internal flow structures through the

use of Particle Image Velocimetry (PIV). Focus will be placed on investigating

Reynolds stresses and coherent structures that develops within the cavity.

2 Experimental details

The experiments were carried out in an open loop wind tunnel with test section

size of 220cm (length) by 40cm (width) by 40cm (height). The turbulence

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 61

tunnel varies from 0m/s to 30m/s. The wind tunnel is represented schematically

in Figure 2. The length of the cavity model is adjustable; L/D of 4 is used. Figure

3 shows the dimensions of the cavity structure. The cavity floor was

instrumented with taps for pressure measurements along the centerline of the

cavity floor. The width of the cavity spans the entire test section. The origin is

taken to be at the leading edge of the cavity and at the middle of the cavity span.

The PIV system used consists of a double pulsed Nd:YAG laser, a HiSense

MkII (Dantec Dynamics) camera and a system hub (FlowMap). The time

between the laser pulses was set to be 50μs. The two images, captured during the

laser pulses, were adaptive correlated successively starting from an initial

interrogation window size of 64 by 64 pixels to a final size of 16 by 16 pixels

with a 50% overlap ratio. A total of 900 time-independent images were captured

in this study. The position of the PIV laser sheet is shown in Figure 4. The plane

of interest lies on the x-y plane along the centreline of the cavity.

3 Experimental results

3.1 Boundary layer and surface pressure measurements

The boundary layer at the leading edge of the cavity was found to conform to the

1/7 power law and hence is deduced to be turbulent. The boundary layer

thickness, δ0.95, was found to be about 90mm. The large thickness could be

attributed to both the blockage effect and the design of the leading plate before

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

62 Advances in Fluid Mechanics IX

the cavity. The pressure coefficient distribution along the centerline of the cavity

floor is shown in Figure 5. The pressure gradient is slightly negative up to a x/L

of ~ 0.4. After which, there is a gradual increase of pressure towards the cavity

aft. These observations are typical of an open flow configuration.

0.4

0.2

Cp

0

0 0.2 0.4 0.6 0.8 1

-0.2

x/L

The 900 independent images acquired from PIV were first averaged to obtain

mean field variables. These included the mean flow field in the x and y direction,

the in-plane streamlines, velocity fluctuations in the x and y direction, Reynolds

shear stress, vorticity and spatial correlation. These quantities will be discussed

below. Figure 6 shows the mean velocity in the x and y direction, U and V, as

well as the in-plane streamlines. The contour distributions, as well as the large

recirculating bubble within the cavity are typical of an open cavity configuration.

′ ′

Normalised velocity fluctuations for the cavity in x and y direction, and

∞ ∞

′ ′

are shown in Figure 7 while the Reynolds shear stress, , is shown in Figure 8.

∞

In each figure, a similar study done by Ukeiley et al. [5] for L/D=5.16 at a Mach

number of 0.2 is shown on the right for comparison. The general distribution of

′

for this study is similar to that of Ukeiley et al. [5] as shown in Figure 7.

∞

However, there is a region of high intensity above the cavity, especially at the

top right corner of the contour plot of this study. This could be due to the thick

boundary layer. If this region is ignored, a maximum value of about 0.3 is found

along the lip line towards the trailing edge of the cavity. This is comparable to

the maximum value of intensity found by Ukeiley et al. [5], which also occurs

′ ′

near the trailing edge of the cavity. Similar to the plots of , the two plots of

∞ ∞

have the same distribution except for the patch of high intensity above the cavity

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 63

for this study. A maximum intensity of 0.2 is found for this study, which is

comparable to the maximum value of 0.170 found by Ukeiley et al. [5]. The

position of the maximum intensity occurs near the trailing wall of the cavity in

both cases. From Figure 8, the Reynolds shear stress distribution of this study is

also similar to that of Ukeiley et al. [5], with the maximum value occurring

towards the trailing edge of the cavity. A maximum value of around 0.03 is

obtained in this study while a maximum value of about 0.015 was obtained by

Ukeiley et al. [5].

Figure 7: Contour plot of and (left: current study, right: Ukeiley et al.

[5]).

Figure 8: Contour plot of (left: current study, right: Ukeiley et al. [5]).

ω

Contour plot of normalized vorticity, (D and U∞, denotes respectively the

∞

cavity depth and free stream velocity) was compared against that of Ukeiley et al.

[5] in Figure 9. Region of higher vorticity was limited to the shear layer in both

cases, with some levels of vorticity spreading towards the aft section of the

cavity. This is due to the high speed free-stream fluid interacting with the low

speed fluid inside the cavity. Both plots have vorticity values in similar range.

Figure 9: Contour plot of (left: current study, right: Ukeiley et al. [5]).

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

64 Advances in Fluid Mechanics IX

the lip line of the cavity, two point spatial correlation coefficients of both the x

and y velocity fluctuations were examined. The spatial correlation in x direction

is represented by Ruu and the spatial correlation in y direction by Rvv. The spatial

correlation for v' velocity is calculated as

′ ′

R x ,y ′ ′

(1)

where x and y denote the reference position for x and y respectively. For this

calculation, a reference point was first chosen. The velocity fluctuation at all

points in an individual image was correlated to the velocity fluctuation at this

reference point. The same correlation procedure was carried out for all 900

images, the result of which was averaged to give the final image. Five reference

positions were chosen for both Ruu and Rvv. These 5 positions lie on the lip line

of the cavity and correspond to positions where x/L=0.1, 0.3, 0.5, 0.7 and 0.9.

The aim of selecting these 5 positions was to track the growth of the shear layer

across the lip. The results are presented in Figure 10, where the lengths of the

structures were also measured and shown in cases when structures are smaller

than the size of cavity.

75mm 40mm

94mm 65mm

136mm 94mm

107mm

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 65

In Ruu, the length and height of the region of high positive correlation grows

with increasing distance from the cavity leading edge. This is consistent with the

amplification of disturbances by the shear layer to form large scale coherent

structures. In Rvv, the correlated regions have roughly circular contours. There

appears to be a distinct pattern of alternating regions of positive and negative

coefficient values across the span of the cavity, shown by the red and dark blue

regions. This reflects the organized nature of the coherent structures within the

cavity. Again, the area of positive correlation grows across the cavity.

The velocity profiles across the shear layer as a function of x/L is shown in

Figure 11. The plots indicate that the shear layer attains self-similarity beyond

x/L=0.3. The growth in momentum thickness as shown in Figure 12 was

estimated to be around 0.031. This is close to the entrainment rate of the mixing

θ

layer measured by Liepmann et al [6] which has value of 0.035.

3 x/L=0.1

x/L=0.2

2

x/L=0.3

1 x/L=0.4

y/x

x/L=0.5

0 x/L=0.6

-0.2 0 0.2 0.4 0.6 0.8 1 1.2 x/L=0.7

-1

x/L=0.8

-2 x/L=0.9

u/U∞

0.01

y = 0.0313x + 0.0024

θ(x)

0.005

0

0 0.05 0.1 0.15 0.2

x

Figure 12: Growth of shear layer across the cavity lip line.

POD was first introduced in the context of turbulence by Lumley [7] as an

unbiased way of extracting coherent structures from turbulent flows. It provides

an empirical basis for representations of complex spatio-temporal fields that is

optimal in the sense that it converges faster on average than any other

representation. Detailed explanations of POD can be found in Holmes et al. [8]

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

66 Advances in Fluid Mechanics IX

and Delville et al. [9]. Sirovich [10] introduced the snapshot POD method that

utilized spatially resolved but time un-correlated snapshots of the flow field like

those obtained by PIV. The spatial modes associated with snapshot POD, ϕ,

were assumed to have a special form in terms of the original velocity data:

ϕ x ∑ a U x (2)

consequential eigenvalue problem is hence deduced to solve for a :

n

Ca n λna (3)

Flow reconstruction can then be done through the linear combination of spatial

eigenfunctions with the temporal eigenfunctions as coefficients. Through POD,

the spatial eigenfunctions are orthonormal and the temporal coefficients are

orthogonal.

U x ∑ a t ϕ x (4)

POD was applied to all 900 images from PIV, with the mean flow field

subtracted from the images before decomposition. Figure 13 is a plot depicting

the individual energy content of the modes, the sum of which contributes to 75%

of the total energy in the flow. A total of about 74 modes were needed to make

up 75% of the energy content in the field. The first mode has about 15.7% of the

total energy. This is followed by a drop in energy for subsequent modes. Mode 2,

3 and 4 has about 7.3%, 5.6% and 4.5% of total energy respectively. For

reconstruction of flow, the first 4, 7 and 14 modes are needed to recover 33%,

40% and 50% of the total energy content respectively.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 67

The first 5 spatial modes in the y direction are shown in Figure 14. These

plots reveal the structures that are contributing to each mode. From the spatial

modes, the length scales of the structures were estimated, where clear structures

could be identified. This is indicated by the white arrows in modes 2 to 5. From

the figure, mode 1 has a structure that stretches through the entire cavity. Hence,

the length-scale is taken to be the length of the cavity, 160mm. Mode 2 and

mode 3 both have a length scale of about 114mm while mode 4 has a length

scale of about 100mm and mode 5 has a length scale of about 83mm. The

decrease in the length scale of structure with the increase in the mode number is

expected as smaller structures contribute to a lesser percent of the total energy in

the flow.

These streamlines represent the contribution of the modes in both the x and y

direction. Inspection of the in-plane streamlines seems to indicate the flapping of

the shear layer. One could deduce this fact from the super position of these

modes onto the mean in-plane streamlines profile as shown in Figure 6 and eqn

(4). Mode one and two indicated a flow structure of the closed cavity type while

mode three indicated the possibility of a large structure ejecting from the aft of

the cavity.

Reconstruction of flow was done with the first 4 modes to see how well these

modes can estimate the original mean flow field as well as whether the turbulent

noise due to the thick boundary layer can be decoupled. The mean quantities are

recalculated from the reconstructed field. Figure 16 shows the mean velocity in

the x and y direction, U and V, as well as the in-plane streamlines. When

comparing Figure 16 with Figure 6, it was observed that both U and V, and

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

68 Advances in Fluid Mechanics IX

Reconstructed normalised velocity fluctuations in x and y direction, and ,

are shown in Figure 17 while the reconstructed Reynolds shear stress, , is

shown in Figure 18. In the contour plot of and , it was observed that the

region of high intensity above the cavity that was present in Figure 7 has now

disappeared. This implies that we could decouple the turbulent background noise

from the results. The general distribution of velocity fluctuations here is about

′

the same as the original flow field in both x and y direction. For , a maximum

∞

value of 0.25 was obtained, which is comparable to the original value of 0.3. For

′

, a maximum intensity of about 0.14 was obtained, which is slightly lower than

∞

the original value of 0.2. This loss of energy was expected as the background

u′v′

turbulent noise has been filtered by the POD procedure. In the contour plot of 2 ,

U∞

the maximum value of 0.03 and its position is similar to the original flow shown

in Figure 8. However, the intensity at the leading edge is lower here than in the

original flow.

streamlines (right), using the first 4 modes.

Figure 17: Contour plot of (left) and (right), modelled by first 4 modes.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 69

Overall, it seems that the first 4 modes were able to replicate the mean flow

field to a fairly accurate extent. This is an important feature of POD since using

less number of modes to model the flow field accurately means that cost of

computation can be reduced or that flow control studies could be carried out

more efficiently.

4 Conclusion

The present work has made used of several data reduction tools to post process

PIV data of an open cavity with L/D=4. The shear layer which grows across the

cavity was found to attain self-similarity in the velocity profiles beyond x/L=0.3.

The rate of growth in momentum thickness of this shear layer was estimated to

be around 0.031, a value close to the entrainment rate of a turbulent mixing

layer. Time-averaged quantities of the flow revealed the mean flow

characteristics within the cavity while spatial correlations of velocity fluctuations

indicated the presence of organized coherent structures which grew across the

shear layer of the cavity. POD was applied to PIV data to derive the structures

responsible for the different modes of flow. From the POD results, the first four

modes were used to reconstruct the flow. This filtering process could remove the

turbulent background noise successfully.

References

[1] Kook, H., Mongeau, L. , Brown, D.V. & Zorea, S.I., Analysis of the interior

pressure oscillations induced by flow over vehicle openings. Noise Control

Engineering Journal, 45, pp. 223-234, 1997.

[2] Gharib, M. & Roshko, A., The effect of flow oscillations on cavity drag.

Journal of Fluid Mechanics, 177, pp. 501-30, 1987.

[3] Rockwell, D. & Naudascher, E., Review-self-sustaining oscillations of flow

past cavities. Transactions of the ASME. Journal of Fluids Engineering,

100, pp. 152-65, 1978.

[4] Plentovich, E.B., Experimental Cavity Pressure Measurements at Subsonic

and Transonic Speeds. NASA Technical Paper 3358. 1993.

[5] Ukeiley, L. & Murray, N., Velocity and surface pressure measurements in

an open cavity. Experiments in Fluids, 38(5), pp. 656-671, 2005.

[6] Liepmann, H.W. & Laufer, J., Investigation of free turbulent mixing,

National Advisory Committee for Aeronautics: Washington, DC, United

States , 1947.

[7] Lumley, J., The Structure of Inhomogeneous Turbulent Flows. Atmospheric

Turbulence and Wave Propagation, Moscow: Nauka, 1967.

[8] Holmes, P., Lumley, J. & Berkooz, G., Turbulence, Coherent Structures,

Dynamical Systems and Symmetry, Cambridge, England: Cambridge

University Press, 1996.

[9] Delville, J., Cordier, L. & Bonnet, J.P., Large-Scale-Structure Identification

and Control in Turbulent Shear Flows, Springer Berlin / Heidelberg, pp.

199-273, 1998.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

70 Advances in Fluid Mechanics IX

Coherent structures. Quarterly of Applied Mathematics, 45, pp. 561-70,

1987.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Section 2

Computational methods

www.engbookspdf.com

This page intentionally left blank

www.engbookspdf.com

Advances in Fluid Mechanics IX 73

T. H. Moulden

The University of Tennessee Space Institute, USA

Abstract

A standard gauge velocity field is introduced into the constant density Navier–

Stokes equations and extended to treat fluid turbulence. Some of the elementary

properties of such gauge fields are explored and relationships between them

established. In particular, the gauge fields separate out the irrotational motion

from the vortical motion. An example that describes the inactive component of

a turbulent flow is presented.

Keywords: gauge fields, turbulence models, energy estimates.

1 Background

Write the field equations for the flow of a constant density viscous fluid in the

standard form:

∂vi ∂vi ∂ ∂P ∂ 2 vi

= 0; + (vi vj ) + =ν + fi (1a, b)

∂xi ∂t ∂xj ∂xi ∂xj ∂xj

where P (x, t) is the pressure field normalized by the constant fluid density.

Equations (1a,b) define the field (v, P ) in which v(x, t) is the instantaneous

velocity field at location x and time t. f(x) denotes a time independent body

force in equation (1b). Herein, the kinematic viscosity coefficient, ν, is taken to

be constant. To be useful these equations must possess a unique regular solution

for meaningful boundary conditions; and the existence of such a solution is

assumed in the following discussion. Certain additional assumptions must be made

before equations (1a,b) can be adopted to describe the turbulent flow of a viscous

fluid. These assumptions include the need for the turbulent velocity and pressure

fluctuations to be continuous and have bounded energy. It is also required that all

velocity-related correlation functions be bounded. As is standard, the divergence of

equation (1b) produces a Poisson equation to compute the pressure field directly:

∂2P ∂fi ∂2

= − (vi vj ) (1c)

∂xi ∂xi ∂xi ∂xi ∂xj

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120071

www.engbookspdf.com

74 Advances in Fluid Mechanics IX

since, as stated above, it is assumed that all required derivatives exist. Some

regularity results for the Navier–Stokes equations are given in Seregin [1].

Let the Reynolds decomposition be expressed in the standard form:

v → V + u; V = E(v) and E(u) = 0 (2)

if u(x, t) is the fluctuating velocity field and V(x, t) the corresponding mean

velocity. Equation (2) implies that the kinetic tensor, R = v ⊗ v will be

decomposed as R → R + R in the mean motion equations provided that

R = V ⊗ V denotes the mean kinetic tensor. R(·) in the above represents an

appropriate mean value operator whose specific form is not of significance for the

present discussion and R = E(u ⊗ u) is referred to as the Reynolds tensor. As is

well known, introduction of the Reynolds decomposition renders equations (1a,b)

unclosed and hence unsolvable without an additional statement in the form of a

turbulence model: χ(R, V, α) = 0 (with a vector, α, of scalar constants). The

constraints that must be placed upon such models, are not the present interest —

for some of which see the discussion in Moulden [2]. Details of specific turbulence

models are given in such references as Chen and Jaw [3], Wilcox [4] as well as in

Launder and Sandham [5]. See also the discussion in Gatski [6].

Start by noting some fundamental properties of equations (1a,b) that will be

required for subsequent developments:

a) As is standard, equations (1a,b), being based upon Newtonian mechanics, are

covariant under the Galilean group, G a :

x∗ = Q[x + VT t + x0 ]; t ∗ = t + t0

which implies a velocity transformation of the form v∗ = Q [v + V T ]. Here,

Q ∈ SO 3 is a constant coordinate rotation while VT is the constant Galilean boost

velocity while x0 and t0 are constant space and time translations. The pressure

field is invariant across inertial frames. Finally, note that the spatial and temporal

gradients transform as:

≡ Qij ; = − (VT )k

∂x∗i ∂xj ∂t∗ ∂t ∂xk

under the group G a . The Reynolds averaged equations, like the Navier–Stokes

equations, are fully covariant under the Galilean group.

b) Gurtin [7] gave a simple uniqueness result for the constant density Navier–

Stokes equations (1a,b) under the assumption that certain regularity conditions are

satisfied. In particular, that the norm of the stretching tensor D = sym(L) (if

L = ∇(v) is the velocity gradient tensor) must be bounded for all x, t of interest.

This constraint implies that D must have bounded eigenvalues: a condition related

to the requirements on the boundedness of the Reynolds tensor R. The theorem

of Gurtin [7] showed that the velocity field, v(x, t), must be unique but that the

pressure field has the uniqueness property only up to an arbitrary function of time:

P (x, t) → P (x, t) + P ∗ (t)

say. This finding is nothing more than noting that the simple equality:

∂ ∂

P (x, t) = [P (x, t) + P ∗ (t)]

∂x ∂x

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 75

must hold for any bounded function P ∗ = P ∗ (t), only. Most important for

practical applications is the velocity field uniqueness obtained from the theorem

of Gurtin [7] and this result will be required herein.

It will be found below that certain gauge transformations introduced in the

literature do not possess full Galilean covariance and so are not, to this extent,

consistent with the Navier–Stokes equations. Some applications of one such gauge

transformation will, however, be explored.

The instantaneous Navier–Stokes equations for constant density flow were

written in equations (1a,b) (and formally constitute a well posed problem

when appropriate boundary and initial conditions are appended: even though

certain issues relating to uniqueness and regularity remain unresolved). The

instantaneous kinetic tensor R = v ⊗ v, again normalized by the fluid density,

has been introduced in equation (1b). Being symmetric the kinetic tensor has real

eigenvalues one of which is λ = v, v, the kinetic energy. It is assumed here

that the velocity field is regular in that all required derivatives exist. The same is

assumed of the pressure field.

With these definition in place the mean motion equations, corresponding to

equations (1a, b), can be written as:

∂V

+ div(R) + div(R) + ∇(P ) = ν∇2 (V) + f; div(V) = 0 (3a, b)

∂t

when the velocity field v(x, t) has been decomposed in the form v → V + u with

|u|| bounded. The corresponding equation for the fluctuating velocity, u(x, t), with

div(u) ≡ 0, is given as:

∂u

+ div[(V ⊗ u + u ⊗ V) + u ⊗ u − R] + ∇(p ) = ν∇2 (u) (4)

∂t

since the body force has been restrained as f = f(x) only. Here the pressure field

is given the usual decomposition as P = P + p . The standard evolution equation

for the Reynolds tensor R = (Rij ) can be established from equation (4) and has

the form:

∂Rij ∂Rij ∂Vi ∂Vj ∂ 2 Rij

+ Vk + Rjk + Rik −ν = −Ψij (5)

∂t ∂xk ∂xk ∂xk ∂x2k

∂ui ∂uj ∂

Ψij = 2νE + E(ui uj uk ) + ψij ≡ ET |ij + DF |ij + ψij

∂xk ∂xk ∂xk

has been included in equation (5). In this equation ET denotes the turbulence

dissipation and DF the velocity diffusion term whose explicit form is not of

importance in the present context. As usual, turbulence models are required for

both of these terms. Finally ψij denotes the standard pressure fluctuation terms in

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

76 Advances in Fluid Mechanics IX

∂(uj p ) ∂(ui p ) ∂uj ∂ui

ψij = E + −E p + (5a)

∂xi ∂xj ∂xi ∂xj

The velocity, v(x, t), is unique and, for a given mean value operator, E, so is

the mean motion V. Hence the fluctuations u(x, t), must also be unique along

with the Reynolds tensor R = E(u ⊗ u) again when determined from the unique

instantaneous velocity field defined by the Navier–Stokes equations. Under the

group G a the well known transformations:

V∗ = Q[V + VT ]; u∗ = Qu

hold (see Speziale [11] for a more general discussion of the invariance properties

of turbulence models). Only the mean velocity is influenced by the Galilean boost

velocity VT . In other words, the turbulent fluctuations are transparent to uniform

rectilinear motion of the inertial frame.

3 A gauge decomposition

The experimental work of Bradshaw [10] studied the inactive component of

the motion external to a turbulent shear layer. This motion contains irrotational

velocity fluctuations as well as pressure fluctuations and the two relate via a

Bernoulli-type equation. It is appropriate to introduce a gauge field ω(x, t) that

describes the irrotational motion. That is equation (1a), wherein div(v) = 0,

allows the specification of a gauge velocity field ω in the form: v → s + ω when:

div(v) = 0 ≡ div(s) + div(ω) (1d)

Since v is a function of space and time, it is assumed that s and ω have that same

dependence. The nature of the gauge field ω(x, t) is arbitrary within that constraint

and could be stochastic. At this point there is nothing to specify that the above

decomposition of the velocity field is unique. However, the variables s and ω can

at most, from equation (1d), differ by a divergence-free vector field. Clearly, when

ω = ∇(ϕ), for some scalar ϕ, the equality ζ = curl(v) ≡ curl(s) holds so that

the s(x, t) field captures all the vorticity in the flow. Now the condition div(v) = 0

provides the Poisson equation div(s) + ∇2 (ϕ) = 0 and the ω field is potential

but not, in general, harmonic. Since the vector field curl(s) = ζ is known a (non-

unique) velocity field can be generated from it. There is non-uniqueness since any

β ∈ R such that curl(grad(β)) ≡ 0 can be introduced additively. However,

since div(v) = 0 has the physical meaning of mass conservation in constant

density flow, the specification of s(x, t) as the vorticity carrying component in

the decomposition v = s + ∇(ϕ) is unique: the scalar β can be combined with

ϕ. For a discussion of the more specific Stokes–Helmholtz decomposition, which

is not required herein, see Wu et al. [12]. Throughout it is assumed that all vector

fields are bounded and, in unbounded regions, tend to zero in the far field.

For the application in mind in section 5, it is appropriate to take the vector ω

to be potential since it can then represent the irrotational component of a turbulent

flow; both inside and outside of a turbulent shear layer. The appropriate boundary

conditions to be appended to the definition of the s and ω fields will be addressed

in section 5 for a specific flow field.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 77

From equation (1c) a Poisson equation for the pressure can be constructed, in

the usual way, to give:

∂2P ∂fi ∂ ∂

= − (sj + ωj ) (si + ωi )

∂xi ∂xi ∂xi ∂xi ∂xj

once the s(x, t) and ω(x, t) variables are known.

The above gauge decomposition was introduced by Weinan E and Liu [13]

and discussed by Wang and Liu [14] in the context of numerical solutions of

the Navier–Stokes equations. The latter adopted a Crank–Nicolson, or backward

Euler, time discretization to construct a stable finite difference formulation. The

objective herein is to explore the possible use of such gauge fields in turbulent

flow: specifically to isolate the inactive motion from the vortical motion inside

a turbulent shear layer. As noted above, these irrotational velocity fluctuations

exist outside the shear layer and the ω field must also describe this external

motion. Various aspects of the physics of the inactive motion are discussed in

Bradshaw [10, 15] and Phillips [16] when restricted to the case of a statistically

stationary flow field. The present interest lies in exploring the extension to non-

stationary turbulent flows. In particular, evolution equations for the inactive motion

are required but now expressed in terms of the gauge fields.

From equations (1a,b) there is an evolution of the velocity field s(x, t) and a

specification of the gauge ω(x, t) given by the equations:

∂si ∂ωi ∂2ϕ ∂si ∂ ∂ 2 si

=− ≡− ; + [vi vk ] = ν + fi (6a, b)

∂xi ∂xi ∂xi ∂xi ∂t ∂xk ∂xk ∂xk

if the variable ωi ≡ ∂ϕ/∂xi is potential, so that v = s + ∇(ϕ), then the pressure

field P (x, t) is determined retrospectively from the equality:

∂ϕ ∂2ϕ

+P =ν (7)

∂t ∂xk ∂xk

once the gauge field ϕ(x, t) has been determined. The regularity assumptions made

in the theory demand that both quantities, ∂ϕ/∂t and ∇2 (ϕ), are bounded. The

pressure field is then well defined in equation (7). It hardly requires stating that

splitting the velocity v into a rotational and irrotational parts is artificial in the sense

that flow experiments do not allow this separation of the measured velocity field.

If equation (7) is looked upon as a non-homogeneous equation for the potential

ϕ(x, t) then the constraints to be imposed upon the function P (x, t), for a unique

solution to exist, must be specified. From Stakgold [17] a boundedness condition

on P (x, t) is required.

The potential ϕ can be obtained from equation (6a) in terms of a Laplacian

Green’s function when the s(x, t) field is known. If ϕ(x, t) is determined uniquely

from equations (6a,b) then the pressure P (x, t) follows uniquely from equation

(7). The classical vorticity transport equation for constant density fluid motion:

∂ζi ∂ζi ∂vi ∂ 2 ζi ∂fk

+ vj = ζj + ν + ijk ; ζ = curl(v) ≡ curl(s) (V )

∂t ∂xj ∂xj ∂xj ∂xj ∂xj

does not make reference to the fluid pressure. It is consistent with this observation

that equation (7) determines the fluid pressure field from just the irrotational

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

78 Advances in Fluid Mechanics IX

holds under that group.

Consider the Galilean transformation applied to the gauge decomposition

introduced above when it is required that:

s∗ = Q[s + VT ] and ω ∗ = Q ω ⇒ | ω ∗| ≡ |ω||

2

in addition |s∗| 2 ≤ |s|| + |VT | as a bound on the vector s∗ (x∗ , t∗ ): a bound that

must be boost velocity dependent. When ω is potential there is the transformation:

∇∗ (ϕ∗ ) = Q ∇(ϕ) under G a so that |∇∗ (ϕ∗ )|| = |∇(ϕ)|| . It should be noted

that the pair of equations (6b,7) are not Galilean covariant under boosts since the

transformed equation (6b) has the form:

∂si ∂ ∂ωi ∂ 2 si

+ (vi vj ) +(VT )j =ν + fi (Ga)

∂t ∂xj ∂xj ∂xj ∂xj

∂ϕ ∂ϕ ∂2ϕ

−(VT )k +P =ν (Gb)

∂t ∂xk ∂xk ∂xk

terms in brackets in equations (Ga,b) destroying the covariance. The equations

(6b,7) are, however, covariant under Galilean rotations. Equation (6a) is fully

Galilean covariant. Equations (Ga,b) do not feature in the following discussion

but must be recalled when transferring to a moving inertial frame.

Introduce a Reynolds decomposition into equations (6a,b) by specifying that:

s → S + r; ϕ → Ψ + φ

where S = E(s) and Ψ = E(ϕ). The vorticity field ζ(x, t) = curl(s) has two

components: curl(S) and curl(r); the latter associating with the rotational part of

the turbulent fluctuations. Given this decomposition, equation (6a) splits into:

=− ; =− (8a, b)

∂xi ∂xi ∂xi ∂xi ∂xi ∂xi

Similarly equation (7) yields the pair of equations:

∂Ψ ∂2Ψ ∂φ ∂2φ

+P =ν ; + p = ν (9a, b)

∂t ∂xi ∂xi ∂t ∂xi ∂xi

for the mean and fluctuating pressure fields when the potentials φ and Ψ are

known. In addition, there is: E(p v) ≡ E(p r) + E(p ∇(φ)). Hence the component

E(p ∇(φ)) depends only on the potential φ(x, t). The component E(p r), if it

exists, is an interaction of the irrotational pressure fluctuations with the vortical

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 79

∂Si ∂ ∂ 2 Si

+ Rik + Rik = ν + fi

∂t ∂xk ∂xk ∂xk

∂ri ∂ ∂ 2 ri

+ [Vi uk + ui Vk + ui uk − Rik ] = ν

∂t ∂xk ∂xk ∂xk

the field r(x, t) being defined from equation (10b). Here R = E(u ⊗ u), the

original Reynolds tensor, relates directly to correlations of the (r, φ) fluctuations

through the above transformations. Explicitly there is:

R = P + 2E(r ⊗ ∇(φ)) + E(∇(φ) ⊗ ∇(φ)), where P = E(r ⊗ r)

Equation (8b) now gives the potential φ(x, t) uniquely from the Poisson Green’s

function in the form:

∂ri (ξ, t)

φ(x, t) = − G(x, ξ) dV (ξ) + Γ(x, t) (11)

D(ξ) ∂ξi

where Γ(x, t) represents the contribution from the boundary conditions. The

corresponding time derivative can be obtained as:

2

∂φ ∂ ri (ξ, t) ∂Γ(x, t)

=− G(x, ξ) dV (ξ) + (11a)

∂t D(ξ) ∂t ∂ξi ∂t

The domain of integration only extends over the support of the field r(x, t)

and that function must be such that the integrals in equations (11,11a) are

bounded. Given the value of ∂φ/∂t from equation (11a) the fluctuating pressure is

determined directly from equations (8b,9b). In both equations (11) and (11a), the

function r(x, t) is only non-zero in any vortical region, such as turbulence, that is

present in the domain of interest. The ∇(φ) term follows from equation (11) as:

∂φ(x, t) ∂G(x, ξ) ∂rj (ξ, t) ∂Γ(x, t)

=− dV (ξ) + (11b)

∂xi D(ξ) ∂x i ∂ξ j ∂xi

Similar statements hold for the mean motion potential Ψ(x, t) in equations (8a)

and (9a). That is, for example:

∂Si (ξ, t)

Ψ(x, t) = − G(x, ξ) dV (ξ) + Λ(x, t) (11c)

D(ξ) ∂ξi

again with the integral taken over the support of the S(x, t) function. Derivatives

of Ψ(x, t) then follow as above.

The relationship between the different velocity decompositions that have been

introduced above needs to be clarified in order to establish the gauge Reynolds

decomposition equations. From the above definitions:

v = V + u = s + ω ≡ S + r + ∇(Ψ) + ∇(φ)

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

80 Advances in Fluid Mechanics IX

u = r + ∇(φ) and V = S + ∇(Ψ) (12a,b)

to define the fluctuating and mean velocity fields. Associated with these velocity

fields are the fluctuating and mean vorticity fields:

ζ ≡ curl(u) = curl(r) and ζ ≡ curl(V) = curl(S)

such that the r(x, t) and S(x, t) vector fields carry the vorticity of the turbulent

fluctuations and the mean motion respectively. The irrotational motion is described

by the mean, ∇(Ψ), and fluctuating, ∇(φ), potential fields. The vortical motion,

r(x, t) and S(x, t) act as sources for the inactive components φ(x, t) and

Ψ(x, t) in equations 8(a,b). The above Reynolds decomposition calls forth the

transformations:

S∗ = Q [S + VT ]; r∗ = Q r; ∇∗ (Ψ∗ ) = Q ∇(Ψ); ∇∗ (φ∗ ) = Q ∇(φ)

under the Galilean group.

Equation (10b) allows the construction of an equation for the Reynolds gauge

stress Pij = E(ri rj ) associated with the vortical component of the flow. This

requires the evaluation of correlations of the form:

E(u ⊗ r) ≡ E(r ⊗ r) + E(r ⊗ ∇(φ))

wherein the first term will, generally, predominate as discussed by Bradshaw [10].

Specifically, there is:

∂Pij ∂Pij 2 ∂ri ∂rj

+ Vk + Πij = ν∇ (Pij ) − 2νE (13)

∂t ∂xk ∂xk ∂xk

(α)

where Πij = Σα=1,7 Πij , with the individual contributions given as:

(1) ∂Vi ∂Vj (2) ∂2φ ∂2φ

Πij = Pjk + Pik ; Πij = Vk E rj + Vk E ri

xk ∂xk ∂xi ∂xk ∂xj ∂xk

(3) ∂Vi ∂φ ∂Vj ∂φ (4) ∂(ri rj rk )

Πij = E rj + E ri ; Πij =E

∂xk ∂xk ∂xk ∂xk ∂xk

(5) ∂2φ ∂2φ (6) ∂(ri rj ) ∂φ

Πij = E rj rk + ri rk ; Πij = E

∂xi ∂xk ∂xj ∂xk ∂xk ∂xk

(7) ∂2φ ∂2φ ∂φ

Πij = E rj + ri

∂xk ∂xi ∂xk ∂xj ∂xk

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 81

neglected then the quantity Π in equation (13) reduces to: Π = Π(1) + Π(4)

only. Equation (13) then takes on the form:

dPij ∂Vi ∂Vj ∂(ri rj rk ) 2 ∂ri ∂rj

+Pjk +Pik +E = ν∇ Pij −2νE (13a)

dt ∂xk ∂xk ∂xk ∂xk ∂xk

which can be compared to equation (5). It is seen that equation (5) contains the

pressure term ψij (given explicitly in equation (5a) ) which is absent from (13a).

This is due to the structure of the gauge decomposition revealed in equations

(9b) and (10b). The pressure field does, in this gauge formulation, only relate

directly to the irrotational part of the flow field. It can be noted that the linearized

compressible flow equations (see Goldstein [18], for example) contain a similar

splitting of the velocity field but with the possibility of vorticity production in

higher order terms. However, the present discussion is restricted to constant density

motion and, as such, there are no entropy changes to take into account (and no

pressure component in the Beltrami diffusion equation to consider). The limit

ρ → const of the compressible Navier–Stokes equations is not a trivial limit;

see the discussion in Schochet [19] for the special case of a baratropic fluid.

The relationship between the compressible and incompressible flow theories is

not direct.

It is well established that not all velocity fluctuations in a turbulent shear layer are

vortical in nature. As mentioned above these irrotational velocity fluctuations, and

the associated pressure fluctuations, propagate outside the turbulent shear layers.

See the discussion in Bradshaw [10, 15] and Phillips [16]. The present interest

lies in adopting the gauge fields introduced above to discuss the physics of the

inactive motion both inside and outside a turbulent shear layer. Moyal [20] showed

that in wave number space the pressure associates with velocity components that

reside in the span of the wave number vector. A significant noise field arises in

high speed turbulent shear layers as discussed, for example, in Phillips [21] and

Tam [22] (with the latter showing photographs of Mach wave propagation external

to a supersonic jet flow).

Outside the shear layer the equality r ≡ 0 holds, but irrotational fluctuations

do exist in that region where the quantity ∇(φ) is non-zero. Phillips [16] took the

irrotational nature of the inactive motion (when statistically stationary) to imply

that it could be explained in terms of a potential. Hence, under these assumptions,

a solution of a Laplace equation was sought for that potential. The theory of

Phillips [16] gave a 1/x4 decay of the energy E(ui ui ) into the far field that

was confirmed by the experiments reported by Bradshaw [15]. However, such

a formulation does not allow the local time dependent external flow field to be

described. The equations developed above allow a more general description of the

irrotational motion.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

82 Advances in Fluid Mechanics IX

Refer to Figure 1 when the equations for region (A) are given as:

∂Si ∂ ∂ 2 Si

+ [Rik + Rik ] = ν

∂t ∂xk ∂xk ∂xk

∂ri ∂ ∂ 2 ri

+ [Vi uk + ui Vk + ui uk − Rik ] = ν

∂t ∂xk ∂xk ∂xk

∂Ψ ∂2Ψ ∂φ ∂2φ

+P =ν ; + p = ν

∂t ∂xi ∂xi ∂t ∂xi ∂xi

∂ri ∂2φ ∂Si ∂2Ψ

=− ; =−

∂xi ∂xk ∂xk ∂xi ∂xk ∂xk

The body force f(x) has not been included in the present discussion. Equations

(11a,14e) allow the pressure fluctuations in region (A) to be given explicitly as:

2

∂ ri (ξ, t) ∂Γ(x, t)

p (x, t) = −ν div(r) + G(x, ξ) dV (ξ) − (15)

D(ξ) ∂t ∂ξi ∂t

when div(r) is computed from equation (14b). Now it is found that the local

pressure fluctuations in region (A) depend upon the local value of div(r) and

an integral over the domain defined by the support of r. In addition, there is in

equation (15) a local contribution from the boundary conditions. This contribution

will vanish for steady flow. The mean pressure, both inside and outside the shear

layer, follows in the same way from equation (9a) when the gauge function Ψ(x, t)

is determined from equation (8a) and the function S(x, t) from equation (10a).

Equations for region (B):

The mean external flow is assumed to lack a vortical component so that the (B)

region flow is defined by the system:

V = ∇(Ψ); u = ∇(φ)

2

∂Ψ ∂ Ψ ∂φ ∂2φ

+P =ν ; + p = ν

∂t ∂xi ∂xi ∂t ∂xi ∂xi

∂2φ ∂Si ∂2Ψ

0=− ; =−

∂xk ∂xk ∂xi ∂xk ∂xk

Note that equations (14c,d) and (16c,d) are identical so that boundary conditions

∇(φ) = 0 are specified at x2 = 0 and in the limit x2 → ∞ and a smooth

solution can be sought on the whole domain. Equations (14d) and (16d) determine

the fluctuating pressure field both inside, and external to, the turbulent shear layer

once the potential φ(x, t) has been determined. At the surface x2 = 0 there is

v ≡ 0 which implies that both equalities V = 0 and u = 0 must hold. It further

follows that r = 0 while ∇(φ) = 0 is also true on this surface. The condition

r = 0 also holds at the edge of the shear layer x2 = L. It is assumed that all

required fields are specified at the upstream boundary x1 = 0 (which is taken to

be some convenient origin for the computation as shown on Figure 1).

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 83

Equation (15) also defines the pressure fluctuations in the region external to the

shear layer where, however, the local div(r) term is identically zero. Alternatively,

equation (16d,16e) define p = −∂φ/∂t directly when equation (11a) applies. The

integration is still over the support of the rotational motion but now the (x, t) values

refer to this external region. For steady boundary conditions, the intensity of the

pressure fluctuations external to the shear layer is given explicitly as:

2

∂ ri ∂ 2 rj

E(p p ) = G(x, ξ) G(x, η) E dV (η) dV (ξ)

D D ∂t ∂ξi ∂t ∂ηj

and a similar expression for the velocity correlation can be obtained from equation

(11b). That is, without the contribution for the boundary conditions the correlation:

∂φ ∂φ ∂G(x, ξ) ∂G(x, η) ∂rk ∂rm

E = E dV (η) dV (ξ)

∂xi ∂xj D D ∂xi ∂xj ∂ξk ∂ηm

is obtained which, from the form of the Green’s function, varies like 1/x42 for

large distances from the shear layer. Pressure velocity correlations follow in the

same way and need not be written down.

6 Final remarks

It has been shown that a non-Galilean invariant gauge velocity field can be

defined for the Navier–Stokes equations when the fluid density is constant.

These gauge fields are obtained from well defined boundary value problems

and provide additional insight into the structure of the constant density Navier–

Stokes equations. Of course, there is greater algebraic complexity in the gauge

formulation. The decomposition has been illustrated with a discussion of the

inactive motion outside a turbulent shear layer.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

84 Advances in Fluid Mechanics IX

References

[1] Seregin, G. Local regularity theory of the Navier Stokes equations. Handbook

of mathematical fluid dynamics, IV, Friedlander, S. and Serre, D. (Eds)., pp

159–200, Elsevier, Amsterdam, 2007.

[2] Moulden, T. H. On properties of turbulence models. Advances in Fluid

Mechanics VIII, Algarve, Portugal, pp 15–26, 2010.

[3] Chen, C-J and Jaw, S-Y. Fundamentals of turbulence modeling. Taylor and

Francis, Washington, 1998.

[4] Wilcox, D. C. Turbulence modeling for CFD. 2nd Edition, DCW, LaCañada,

1998.

[5] Launder, B. E. and Sandham, N. D. (Eds) Closure Strategies for Turbulent

and Transitional Flows. Cambridge University Press, Cambridge, 2002.

[6] Gatski, T. B. Constitutive equations for turbulent flows. Th. and Comp. Fluid

Dyn. 18, pp 345–369, 2004.

[7] Gurtin, M. E. An introduction to continuum mechanics. Academic Press, New

York, 1981.

[8] Frost, W. and Moulden, T. H. (Eds) Handbook of Turbulence. Plenum Press,

New York, 1977.

[9] Moulden, T. H. A Uniqueness Theorem in Fluid Turbulence. Proc. 8th

HEFAT Conference, Mauritius, 2011.

[10] Bradshaw, P. Irrotational fluctuations near a turbulent boundary layer. JFM,

27, pp 209–230, 1967.

[11] Speziale, C. G. Invariance of Turbulent Closure Models. Phys. Fluids, 22 pp

1033–1037, 1979.

[12] Wu, J-Z, Ma, H-Y and Zhou, M-D. Vorticity and Vortex Dynamics. Springer,

Berlin, 2006.

[13] Weinan E and Liu, J-G. Finite difference schemes for incompressible flows in

velocity-impulse density formulation. J. Comp. Phys. 130, pp 67–76, 1997.

[14] Wang, C. and Liu, J-G. Convergence of Gauge Method for Incompressible

Flow. Math. Comp. 69, pp 1385–1407, 2000.

[15] Bradshaw, P. ‘Inactive’ motion and pressure fluctuations in turbulent

boundary layers. JFM. 30, pp 241–258, 1967.

[16] Phillips, O. M. The irrotational motion outside a free turbulent boundary.

Proc. Camb. Phil. Soc., 51, pp 220–229, 1955.

[17] Stakgold, I. Green’s functions and boundary value problems. Wiley, New

York, 1979.

[18] Goldstein, M. E. Aeroacoustics. McGraw Hill, New York, 1976.

[19] Schochet, S. The mathematical theory of the incompressible limit in fluid

dynamics. Handbook of mathematical fluid dynamics, IV, Friedlander, S. and

Serre, D. (Eds)., pp 159–200, Elsevier, Amsterdam, 2007.

[20] Moyal, J. E. The spectra of turbulence in a compressible fluid: eddy

turbulence and random noise. Proc. Camb. Phil. Soc. 48, pp 329–344, 1952.

[21] Phillips, O. M. On the generation of sound by supersonic turbulent shear

layers. JFM. 9, pp 1–28, 1960.

[22] Tam, C. K. W. On the noise of a nearly ideally expanded super sonic jet JFM.

51, pp 69–95, 1972.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 85

incompressible viscous flow simulation

Z. Qian1,2 & C.-H. Lee3

1

AVIC Aerodynamics Research Institute, China

2

AVIC Aeronautical Science and Technology Key Lab for High Speed

and High Reynolds Number Aerodynamic Force Research, China

3

National Laboratory for Computational Fluid Dynamics,

School of Aeronautic Science and Engineering, Beihang University,

China

Abstract

A new HLLC approximate Riemann solver with preconditioning technique based

on the pseudo-compressibility formulation for numerical simulation of the

incompressible viscous flows has been proposed, which follows the HLLC

Riemann solver (Harten, Lax and van Leer solver with contact resolution

modified by Toro) for the compressible flow system. In the authors’ previous

work, the preconditioned Roe’s Riemann solver was applied to the finite

difference discretization of the inviscid flux. Although the Roe’s Riemann solver

is found to be an accurate and robust scheme in various numerical computations,

the HLLC Riemann solver is more suitable for the pseudo-compressible Navier-

Stokes equations, in which the inviscid flux vector is a non-homogeneous

function of degree one of the flow field vector; however the Roe’s solver is

restricted to the homogeneous problems. Numerical investigations have been

performed in order to demonstrate the efficiency and accuracy of the present

procedure. The present results are found to be good agreement with the exact

solutions, existing numerical results or experimental data.

Keywords: precondition, HLLC scheme, pseudo-compressibility, incompressible

viscous flows, LU-SGS.

1 Introduction

The difficulty of numerically solving the incompressible Navier-Stokes (N-S)

equations is the decoupling of the velocity and pressure fields while the zero

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120081

www.engbookspdf.com

86 Advances in Fluid Mechanics IX

proposed by Chorin [1] introduced an additional time derivative of the pressure

field to the continuity equation, and so the continuity and momentum equations

are solved in a coupling manner.

The precondition concept applied on the incompressible flow simulation was

proposed by Turkel [2], then the authors [3] further developed this procedure and

applied it to numerical simulation of the pseudo-compressible N-S equations.

The influences of precondition parameter and pseudo-compressibility factors on

the convergence rate were also investigated systematically, and the optimal

parameters were identified in the author’s previous paper [3].

Although the pseudo-compressibility method has been widely used for about

three decades [4–7], only two classes of scheme, namely the Jameson’s central

scheme [8] and the Roe’s upwind scheme [9], are employed for the inviscid flux

terms. The author’s previous work [3] shown that the Roe’s upwind scheme is

superior to the Jameson’s central scheme for lower dissipation and free of turned

parameter. However, since the inviscid flux vector F of the pseudo-compressible

N-S equations is not homogeneous function of degree one of the flow field

vector U , namely F AU , and also the homogeneous property is a critical

restriction in Roe’s solver, consequently Roe’s upwind scheme may be not

proper in the pseudo-compressibility case.

In the present paper, a new approximate Riemann solver for upwind scheme

for incompressible flow is derived following the HLL scheme propose by Harten

et al. [10] and its modified form HLLC scheme proposed by Toro et al. [11] for

compressible flow simulations. In this approximate Riemann solver we do not

assume the homogeneous property of the inviscid flux, so the proposed

approximate Riemann solver is more proper and general for incompressible flow

simulation.

A numerical procedure using both the proposed HLLC upwind finite

difference scheme and the precondition technology is then established for the

viscous incompressible flow simulation. Several numerical test cases, including

low Reynolds number flow over a circular cylinder, laminar flow over a flat

plate, and high Reynolds turbulent flow over the S809 airfoil with the two

equations k-ω SST turbulent model, are numerically investigated in this paper in

order to demonstrate the efficiency and accuracy of the present procedure. The

present results are found to be good agreement with the exact solutions, existing

numerical results or experimental data.

equations

The pseudo-compressibility procedure and the preconditioning technique for

convergence acceleration for stiff hyperbolic equations are employed to develop

a highly efficient in-house code for solving two dimensional incompressible

flows numerically in the author’s previous work [3].

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 87

Cartesian coordinates can be given as below

1Qt ( E E v ) x ( F F v ) y 0 , (1)

where

1 2 0 0

1 u 2 1 0 , (2)

v 2 0 1

v v

The state variables Q and fluxes E , F , E , F are identical to those given in

[3], where denotes the pseudo-compressibility factor, and denotes the

parameter introduced by precondition. The selection of parameters and is

critical to the accuracy and convergence rate, and some detailed discussions are

already given in [3].

3 Numerical schemes

The semi-discretized finite difference method is used in the present paper. First,

the spatial derivatives are discretized while the time variation remains

continuous. Then, the time marching in temporal can be solved by the implicit

LU-SGS (Lower-Upper Symmetric Gauss-Seidel) algorithm.

the HLL scheme propose by Harten, Lax and van Leer and its modified form

HLLC scheme proposed by Toro, Spruce and Speares for compressible flow

simulations.

Harten et al. [10],Toro [12] put forward the following approximate Riemann solver

U L , if x t S L ,

U ( x, t ) U hll , if S L x t S R ,

ˆ (3)

U , if x t S .

R R

where is U hll is the integral average of the exact solution of the Riemann

problem between the slowest and fastest waves, S L and S R are the fastest wave

velocities perturbing the initial data states U L and U R respectively. Figure 1

shows the structure of this approximate solution of the Riemann problem.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

88 Advances in Fluid Mechanics IX

t

SL SR

U hll

UL UR

Provided that the signal speeds S L and S R are known, U hll can be given by

S RU R S LU L FL FR

U hll (4)

SR SL

following Harten, Lax and van Leer’s formula. The corresponding flux at the cell

interface for the HLL scheme is then given by

FL , if 0 S L ,

S F S F S S (U U )

Fi hll1 2 R L L R L R R L

, if S L 0 S R , (5)

S R S L

FR , if 0 S R .

Given an algorithm to compute the speeds S L and S R one can have an

approximate intercell flux. Procedures to estimate the wave speeds S L and S R in

this paper are given by

SL min{uL aL , uR aR } , SR max{uL aL , uR aR } . (6)

Following Harten, Lax and van Leer, the HLL scheme for the pseudo-

compressibility Navier-Stokes equations can be derived. The artificial speed of

sound is given by

aL , R uL2, R 2 . (7)

For incompressible flow the density is assumed to be constant and the energy

equation is decoupled from the continuity and momentum equations. So the

intermediate pressure p* and normal velocity u* can be obtained from the

Rankine-Hugoniot Conditions across the two wave system similar to the original

HLL scheme for compressible flow, which is given as

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 89

u* , (8)

SR SL

S R p R S L p L 2u L 2u R

p* . (9)

SR SL

Also for incompressible flow, the two wave speeds are always satisfied that

ua 0, u a 0 . (10)

This means that the two approximate wave speeds S L and S R are always

satisfied that

SL 0 , SR 0 , (11)

consequently the corresponding cell interface flux of the HLL scheme for

incompressible can be simplified as

S R FL S L FR S L S R (U R U L )

Fi hll1 2 . (12)

SR SL

Note that this Riemann solver consists of just three constant states separated

by two waves. All intermediate states separated by intermediate waves are

lumped into the single U hll state without regard for the spatial variations of the

solution of the Riemann problem in the Star region, so a shortcoming of the HLL

scheme, with too serious dissipation, is exposed.

As pointed out by Harten et al. themselves [10], the too much dissipation of the

HLL Riemann solver may be alleviated by restoring the missing waves.

Accordingly, Toro et al. [11, 12] proposed the so called HLLC scheme, where C

stands for Contact. In the HLLC scheme the missing middle waves are put back

into the structure of the approximate Riemann solver.

Consider Figure 2, in which the complete structure of the solution of the

Riemann problem is depicted. As shown in the figure, in addition to the slowest

and fastest signal speeds S L and S R , we include a middle wave of speed S * ,

corresponding to the shear wave with eigenvalue u .

Following Toro et al. [11], the HLLC flux can be written as

FL , if 0 S L ,

F F S (U U ), if S L 0 S* ,

(13)

Fi hllc *L L L *L L

if S* 0 S R ,

12

F

*R FR S R (U *R U L ),

FR , if 0 S R .

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

90 Advances in Fluid Mechanics IX

t S*

SL U*L SR

U *R

UL UR

where U *L and U *R denote to the integral averages between the shear wave and

the left and right waves respectively, which can be given by the following

Rankine-Hugoniot conditions across each of the waves of speeds S L , S * and

SR .

Here condition (14)–(16) are three equations for the four unknowns vectors U * L ,

U *R , F*L and F*R . The aim is to find the vectors U *L and U *R , so that the

fluxes F*L and F*R can be determined from (14) and (16) respectively. For

incompressible flow the following conditions are must guaranteed.

Consequently, we can obtain that u* and p* are the same as equation (8) and

(9) respectively, and v*L v L , v*R v R , S * u* . Since S L 0 and S R 0 ,

the corresponding intercell flux of the HLLC scheme for incompressible can be

simplified as

FL* FL S L (U * L U L ), if 0 S* ,

12

Fi hllc (18)

FR* FR S R (U *R U L ), if S* 0.

In order to promote the accuracy, we employ the second order MUSCL scheme

[13] in the present study. Accordingly, the left and right states are given

respectively by

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 91

U L U i L U i ,

(19)

U R U i 1 R U i 1 ,

paper, the minmod limiter is used unless specified.

3.1.4 Viscous term computations and numerical boundary conditions

The viscous terms of eq. (1) are discretized by the second-order central

difference. The boundary conditions imposed on the solid wall are the no-slip

condition for viscous flows, and the pressure on the wall is evaluated by

extrapolation from the interior points. For the far field boundary, the conditions

are specified according to the direction of the in-flow or out-flow, where the

velocities are given, and the pressure is extrapolated from the interior nodes at

the inflow boundary; whereas the pressure is given, and the velocities are

extrapolated from the interior nodes at the outflow boundary.

by Yoon and Jameson [14] is used, which is one of the most commonly used

implicit methods for compressible flows simulations. We generalize the

formulation to the preconditioned pseudo-compressible Navier-Stokes equations,

and details can be found in the authors’ previous work [3].

4 Numerical results

Some validating tests are presented in this section, including low Reynolds

number flow over a circular cylinder, laminar flow over a flat plate, and high

Reynolds turbulent flow over the S809 airfoil with the two equations k-ω SST

turbulence model. These cases are chosen for comparisons with exact analytic

solutions or existing experimental data.

benchmark problem for incompressible flows, for which many experimental and

computational results can be utilized for comparison [4, 15, 16]. In this section,

the problem is simulated by the algorithm proposed in previous sections.

The characteristics of the flow field are sensitive to the Reynolds number.

The case of Re=40 is computed, and the flow renders as a pair of symmetric

separated vortices. The mesh used in this paper consists of a single gird zone

with an O type topology, and is composed of 65×181 in the streamwise and wall

normal direction, respectively, with the first grid spacing of 0.005 diameters and

stretching to the outer boundary of 20 diameters in the normal direction.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

92 Advances in Fluid Mechanics IX

The streamlines of the flow at Re=40 computed by the present scheme are

depicted in Figure 3. The separation length measured from the rear stagnation

point of the cylinder in cylinder diameters, the separation angle which defines

the point of separation from the body, and the drag coefficient in each case are

also shown in Table 1.They are consistent with the experimental data [15, 16]

and the computational results listed in [4]. The errors of the separation length

and separation angle relative to [16] are 2.35% and 0.37% respectively, and that

of the drag coefficient relative to [15] is 2.42%. Moreover, all the errors of the

numerical results relative to the experimental data are less than 5%, which

demonstrate the reliability of the present method.

Re=40.

Length of Angle of

Drag coefficient

separation (L/D) separation (°)

(Errors relative to

(Errors relative to (Errors relative to

[15])

[16]) [16])

Reference [4]

2.29 (+7.51%) 53.0 (-0.93%) 1.549 (-6.12%)

(Numerical)

Reference

[15] — — 1.65

(experimental)

Reference

[16] 2.13 53.5 —

(experimental)

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 93

The incompressible laminar flow over a flat plate is commonly called a Blasius

flow with an analytical similarity solution. This test case is used to validate the

capability of the presently developed procedure for boundary layer flow

simulation. The mesh is composed of 109×50 grids in the stream wise and wall

normal direction, respectively, clustered at the leading and trailing edges of the

flat plate. The computational region is stretched to the upper and upwind

boundary of 5 times the length of the plate, and to the out flow boundary with 6

times the length. The Reynolds number defined by the length of the plate is

1.0 105 . The grid is clustered near the wall due to great gradient in the boundary

layer.

The velocity profile computed by the present scheme is plotted in Figure 4

together with the Blasius solution, where the abscissa y ' y Re x x , and the

ordinate is the dimensionless velocity u . It is found that the present

preconditioned HLLC scheme is capable of providing numerical solutions well

consistent with the Blasius solution.

The S809 airfoil was designed specially for wind turbine applications, the

thickness ratio of which is 21%. A 0.6m chord length model of this airfoil has

been tested in a 1.8m×2.5m low turbulence wind tunnel [17]. In this numerical

test, the Reynolds number is 2.0×106 and the angle of attack is 9.22°. It shows

the reliability of the proposed algorithm for turbulent flow simulation.

The mesh as shown in Figure 5 consists of a single gird zone with an O type

topology, and is composed of 307×120 in the stream wise and wall normal

direction, respectively, with the first grid spacing of 1.0×10-5 chord length and

stretching to the outer boundary of 20 diameters. The Menter’s k SST

turbulence model [18], which takes the advantages of the k model [19] and

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

94 Advances in Fluid Mechanics IX

Wilcox’s k model [20], is used. This model excludes the sensitivity to the

inflow turbulence intensity and retains the numerical stability near the wall.

The computed contours of pressure, turbulence viscosity ratio and turbulence

kinetic intensity are shown in Figure 6. Figure 7 depicts the comparison of the

computed wall pressure coefficient and the experimental test data [17], and good

agreement is shown.

(a) Pressure (b) Turbulence viscosity ratio (c) Turbulence kinetic energy

Figure 6: The contours of pressure, turbulence viscosity ratio and turbulence

kinetic intensity.

coefficients.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 95

5 Conclusions

This paper derives the preconditioned HLLC finite difference scheme for the

incompressible flow by virtue of the pseudo-compressibility procedure. The

proposed scheme does not assume the homogeneous property of the inviscid

flux, so the proposed approximate Riemann solver is more proper and general for

incompressible flow simulation. The preconditioned technique is used to

accelerate the time marching, and the implicit LU-SGS time integration method

is adopted to integrate temporally. The reliability of the present procedure are

demonstrated by applications to low Reynolds number flow over a circular

cylinder, laminar flow over a flat plate, and high Reynolds turbulent flow over

the S809 airfoil. It is found that the numerical results of the present algorithm are

in good agreement with theoretical solutions or experimental data.

References

[1] Chorin A J. A numerical method for solving incompressible viscous flow

problems. J. Comput. Phys., 2: 12–26, 1967.

[2] Turkel E. Preconditioning methods for solving the incompressible and low

speed compressible equations. J. Comput. Phys., 72: 227–298, 1987.

[3] Qian Zhansen, Zhang Jingbai, Li Chunxuan. Preconditioned pseudo-

compressibility methods for incompressible Navier-Stokes equations.

SCIENCE CHINA Phys. Mech. Astronomy., 53(11): 2090-2102, 2010.

[4] Rogers S E, Kwak D. An Upwind-Differencing Scheme for the

Incompressible Navier-Stokes Equations. NASA TM-101051, 1988.

[5] Rogers S E, Kwak D. Upwind differencing scheme for the time-accurate

incompressible Navier-Stokes equations. AIAA Journal, 28: 253–262,

1990.

[6] Kwak D, Chakravarthy S R. A three-dimensional incompressible Navier-

Stokes flow solver using primitive variables. AIAA Journal, 24: 390–396,

1986.

[7] Tamamidis P, Zhang G, Assanis D N. Comparison of pressure-based and

artificial compressibility methods for solving 3D steady incompressible

viscous flows. J Computational Physics, 124: 1–13, 1996.

[8] Jameson A. Time dependent calculations using multigrid with application

to unsteady flows past airfoils and wings. AIAA Paper 91-1596, 1991.

[9] Roe P L. Approximate Riemann Solvers, parameter vectors and difference

schemes. J. Comput. Phys., 43: 357-372, 1981.

[10] Harten A, Lax P D, van Leer B. On upstream differencing and Godunov-

type schemes for hyperbolic conservation laws. SIAM Review, 25(1): 35-

61, 1983.

[11] Toro E F, Spruce M, Speares W. Restoration of the contact surface in the

HLL-Riemann solver. Shock Waves, 4: 25-34, 1994.

[12] E.F.Toro. Riemann solvers and numerical methods for fluid dynamics: A

practical introduction. Springer, 1997.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

96 Advances in Fluid Mechanics IX

second-order sequel to Godunov’s method. Journal of Computational

Physics, 32: 101-136, 1979.

[14] Yoon S, Jameson A. Lower-upper symmetric-Gauss-Seidel method for the

Euler and Navier-Stokes equations. AIAA Journal, 26: 1025– 1026, 1988.

[15] Tritton D J. Experiments on the flow past a circular cylinder at low

Reynolds numbers. J Fluid Mech, 6: 547–567, 1959.

[16] Contanceau M, Bouard R. Experimental determination of the main features

of the viscous flow in the wake of a circular cylinder in uniform translation.

Part 1. steady flow. J Fluid Mech, 79: 231– 256, 1972.

[17] Somers D M. Design and experimental results for the S809 airfoil.

NREL/SR-440-6918, 1997.

[18] Menter F. Zonal two equation k-w turbulence models for aerodynamic

flows. AIAA Paper 93-2906, 1993.

[19] Abid R. Evaluation of two-equation turbulence models for predicting

transitional flows. International journal of Engineering Science, 131: 831-

840, 1993.

[20] Wilcox D. Turbulence modeling for CFD. DCW Industries, Inc., La

Canada, California, 1993.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 97

simulating compressible viscous flow in

complex geometries

B. Jastrow & F. Magagnato

Institute of Fluid Machinery, Karlsruhe Institute of Technology, Germany

Abstract

This paper presents an immersed boundary method combined with a wall-layer

approach implemented into an established flow solver. In the outer flow field,

the compressible Navier-Stokes equations are solved using an approximate

Riemann Solver whereas simplified boundary-layer equations are solved near the

wall. Turbulence is accounted for by the one-equation model of Spalart-Allmaras

in the outer flow region and by a mixing length eddy viscosity model with near

wall damping in the wall layer. Computations performed for various test cases

show good agreement with reference data found in literature.

Keywords: immersed boundary method, wall-layer model, compressible flow.

1 Introduction

Body-fitted grid generation for simulating flow in complex geometry can be very

cumbersome especially when using a block-structured solver. Therefore the

Cartesian-grid immersed boundary method (IBM) offers an interesting approach

since automatic mesh generation can be realized easily. Furthermore, due to

smoothness and orthogonality, Cartesian grids offer high accuracy and

efficiency.

In the IBM a complex geometry is immersed into a regular Cartesian grid.

The effect of the body on the flow is mimicked by the imposition of proper

boundary conditions that act as forcing conditions [1, 2].

Several applications of the IBM use a linear interpolation for setting the

boundary conditions as proposed in [1]. Their validity is therefore only given for

grids resolved down to Δ 1. For flow of high Reynolds number these

restrictions cannot be held in accordance with acceptable computing time. With

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120091

www.engbookspdf.com

98 Advances in Fluid Mechanics IX

the use of a wall-layer model one can overcome the need for high near wall

resolution. Wall models based on incompressible turbulent boundary layer

equations were proposed and tested by Balaras et al. [3], Cabot and Moin [4],

Wang and Moin [5] in a body-fitted context. The applicability of such a wall-

layer model in the framework of the IBM has been studied by Tessicini et al. [6].

Bond and Blottner [7] developed a compressible wall model, named the

diffusion model.

The present paper studies the compressible wall-layer model within the

framework of the IBM. It is shown, that the implementation in the in-house flow

solver SPARC (Structured Parallel Research Code) [8], in the following referred

to as SPARC-IBM, provides results for basic test cases in good agreement with

literature.

2 Mesh generation

The mesh is automatically generated via a ray tracing technique. Based on a

geometry described by a closed surface triangulation every cell center of a

uniform Cartesian grid is marked as either internal or external [9]. Those internal

cells having at least one external neighbor cell are marked as wall-layer cells.

From the latter the normal to the closest wall is computed and stored together

with the forcing point and the appropriate interpolation neighbors (figure 1). The

current implementation of the mesh generator uses a block-refinement algorithm

based on the wall distance.

Figure 1: Mesh in the near wall region: ○: internal cell, Δ: wall-layer cell,

x: external cell, ●: forcing point, i: interpolation neighbors.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 99

3 Numerical method

The Reynolds-averaged Navier-Stokes (RANS) equations for compressible flow

can be written as follows:

0 (1)

0 (2)

0. (3)

For the calculations carried out in this paper, the Spalart-Allmaras turbulence

model has been used for closure. Details of the model are given in the paper by

Spalart and Allmaras [10]. The discretization of the convective terms is obtained

by an approximate Riemann solver (HLLC) [11] whereas the viscous terms are

discretized with a central difference scheme. The solution vector is updated via a

Runge-Kutta scheme.

Inside the wall-layer a simplified set of equations is solved. It is assumed that

convection is negligible compared to diffusion, the normal pressure gradient is

zero and the streamwise gradients are of orders lower than the gradients normal

to the wall. Furthermore, the normal velocity is insignificant compared to the

tangential velocity. This leads to the following equations for the x-momentum

and the energy:

(5)

. (6)

The properties and are the turbulent viscosity and the turbulent

conductivity respectively. In a first stage of implementation, is obtained by a

mixing length eddy viscosity model with near wall damping

1 , (7)

where 0.4, 19 and defines the dimensionless distance to the wall

[5]. At the wall, a no-slip boundary condition is applied. At the forcing point,

that defines the outer edge of the wall-layer, the boundary condition is obtained

through an interpolation from the flow values of the outer flow field. In this

paper, the equations are further simplified by neglecting the time derivative

terms and the streamwise pressure gradient. Two tridiagonal systems are created

for the velocity and the temperature respectively and solved in a segregated way

with the other dependent variables held constant. Subsequently the turbulent

viscosity is updated. An outer loop runs over all equations until the full solution

converges. Finally the flow values for the position of the center of the wall-layer

cell is extracted and provided to the outer flow field as a boundary condition.

Assuming steady-state in the wall-layer corresponds to an instantaneous

response of the wall-layer to the outer flow field which creates some error for

unsteady flow calculations.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

100 Advances in Fluid Mechanics IX

4 Results

4.1 Laminar flow past a flat plate

The first test case is the laminar flow past a flat plate. Since the wall-layer model

also captures the linear near-wall behavior for low Reynolds number flow, this

test case is perfectly suited for verifying the implementation. The computation

was carried out on a structured multi-block mesh with refinement. The boundary

layer was resolved with 32 cells making a total of 54,000 cells in the 2D plane.

Rex=10,000 10,000.

The Mach number was chosen to 0.3 and the velocity profile was

extracted at a Reynolds number of 10,000. In Figure 2 the profile is

compared with the analytical results of the Blasius-equation [12] where ′ is the

dimensionless velocity and is the dimensionless wall distance, defined as

(8)

. (9)

The numerical results show very good agreement with the analytic solution of

Blasius.

The second test case was chosen to prove the consideration of turbulence in the

wall-layer model. The mesh consists of 123,000 cells in the 2D plane which

results in 27 cells resolving the boundary layer. Figures 3 and 4 show the

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 101

obtained velocity profile for 10,000,000 and the skin friction coefficient

along the flat plate together with the experimental data of [13]. The agreement is

quite satisfactory.

Rex=10,000,000.

numbers.

To test the IBM together with curved surfaces, the flow past a circular cylinder

was computed. Two flow regimes are shown, a steady flow at 20 and an

unsteady flow at 200. A local view of the mesh is provided in Figure 5.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

102 Advances in Fluid Mechanics IX

The wake region has been refined manually since a flow adaptive mesh

refinement is subject to future work. The mesh consists of 105,000 cells in the

2D plane. The free stream Mach number in both cases was set to 0.3.

The streamlines for the flow at 20 are shown in Figure 6 whereas Table 1

lists the separation length and angle and the drag coefficient in comparison with

reference values from literature [14, 15]. The agreement is quite satisfactory,

only the separation angle is predicted too low, which might be due to the lack of

considering the pressure gradient in the wall-layer.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 103

for 20.

laminar flow around the cylinder at 200. A comparison of the Strouhal

number and the mean drag coefficient with results from literature [16, 17] is

provided in Table 2.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

104 Advances in Fluid Mechanics IX

provided as a further example. The mesh is taken from the previous stated

calculations and the free stream Mach number was set to 0.3. Figure 8

shows the time averaged pressure distribution in terms of the pressure coefficient

in comparison with experimental results from Cantwell and Coles [18]. The

angle refers to the position on the surface of the cylinder beginning at the

stagnation point. The calculation cannot predict the pressure distribution well.

The minimum is about 40% lower than the measured value. In the

separation area the pressure coefficient is determined too high resulting in a total

drag coefficient of 0.52 which is much lower than the measured value of

1.24. Differences are also obtained for the Strouhal number. The computed

value of 0.23 is much higher than the measured value of 0.18. The

discrepancies are mostly due to the lack of the RANS turbulence model to

predict the separation point correctly. Furthermore, turbulence is treated as

isotropic which does not reflect reality.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 105

5 Post processing

Plotting the flow field obtained by SPARC-IBM requires an additional post

processing since there exists no information about a body-fitted shape of the

wall-layer cells. Therefore the boundary knots of every wall-layer cell are moved

onto the closest wall via a ray tracing technique similar to the one used for the

mesh generation. Figure 9 shows a section of the mesh around a cylinder with

and without the additional post processing. The appropriate flow values are

obtained by an inverse distance interpolation or from the wall values directly if

applicable.

Figure 9: Local view of the mesh around a circular cylinder with (left) and

without (right) the additional IBM-post processing. Every eighth

cell is plotted.

The immersed boundary method (IBM) with a wall-layer approach has been

implemented into an established block-structured code for solving compressible

flow. The computation of basic test cases for laminar flow showed very good

agreement with results found in literature. Viscosity dominated turbulent flow

like the flow past a flat plate was also captured well. Discrepancies were

obtained for turbulent flow with separation exemplarily shown for the flow past

a circular cylinder. However, this is due to the well-known problem of

turbulence modeling of bluff bodies rather than to the IBM.

To overcome the lack of consistency, future work comprises the testing of the

simplified Spalart-Allmaras turbulence model for the wall-layer. It is believed

that in contrast to using the algebraic model, the therewith associated full

coupling of the outer flow field and the wall-layer provides better results for

turbulent flows. Furthermore, the streamwise pressure gradient in the wall-layer

and a flow adapted mesh refinement is about to be implemented.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

106 Advances in Fluid Mechanics IX

Figure 10: Slice through the IBM-mesh of an IC-engine. Every eighth cell is

plotted.

Since the advantages of the IBM are only given for the simulation of complex

geometries where the time for mesh generation is significant, the next step is to

validate the implementation for 3D cases. As an example for a complex mesh a

slice through the mesh of an internal combustion engine is shown in figure 10.

References

[1] Fadlun, E.A., Verzicco, R., Orlandi, P. and Mohd-Yusof, J., Combined

immersed-boundary finite difference methods for three-dimensional

complex flow simulations. J. Comp. Phys., 161, pp. 35-60, 2000.

[2] Mohd-Yosuf, J., Combined immersed-boundary/B-spline methods for

Simulations of flow in complex geometries. Annual Research Briefs,

Center of Turbulence Research, pp. 317-328, 1997.

[3] Balaras, E. et al. Two-layer approximate boundary conditions for large-

eddy simulations. AIAA Journal, 34, pp. 1111-1119, 1996.

[4] Cabot, W., Moin, P., Approximate wall boundary conditions in the large-

eddy simulation of high Reynolds number flow. Flow, Turbulence and

Combustion, 63, pp. 269-291, 1999.

[5] Wang, M., Moin, P., Dynamic wall modeling for LES of complex turbulent

flows. Physics of Fluids, 14, pp. 2043-2051, 2002.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 107

[6] Tessicini, F., Iaccarino, G., Fatica, M., Wang, M., Verzicco, R., Wall

modeling for large-eddy simulation using an immersed boundary method.

Annual Research Briefs, Center of Turbulence Research, pp. 181-187,

2002.

[7] Bond, R.B., Blottner, F.G., Derivation, implementation, and initial testing

of a compressible wall-layer model. Int. J. Numer. Meth. Fluids, 66,

pp. 1183-1206, 2011.

[8] Magagnato, F., KAPPA – Karlsruhe Parallel Program for Aerodynamics,

TASK Quarterly Vol.2, 2, pp. 215-270, 1998.

[9] Verzicco, R., Iaccarino, G., Immersed Boundary Technique for Large-

Eddy-Simulation. Lecture series on “Large Eddy Simulation and related

techniques: Theory and Applications", 2006.

[10] Spalart, P.R., Allmaras, S.R., A one-equation turbulence model for

aerodynamic flows. La Recherche Aerospatiale, 1, pp. 5-21, 1994.

[11] Toro, E.F., Spruce, M., Speares, M., Restoration of the contact surface in

the HLL-Riemann solver, Shock Waves, 4, pp. 25-34, 1994.

[12] Blasius, H., Grenzschichten in Flüssigkeiten mit kleiner Reibung. Z. Math.

Physik, 56, pp. 1-37, 1908.

[13] Wieghardt, K., Tillman, W., On the Turbulent Friction Layer for Rising

Pressure. NACA TM-1314, 1951.

[14] Sucker, D., Brauer, H., Fluiddynamik bei quer angeströmten Zylindern.

Wärme- und Stoffübertragung, 8, pp. 149-158, 1975.

[15] Dennis, S.C.R., Chang, G.Z., Numerical solutions for steady flow past a

circular cylinder at Reynolds numbers up to 100. Journal of Fluid

Mechanics, 42, pp. 471-489, 1970.

[16] Linnick, M.N., Fasel, H.F., A high-order immersed boundary method for

unsteady incompressible flow calculations. J. Comp. Phys., 204, pp. 157-

192, 2005.

[17] Liu, C., Zheng, X., Sung, C.H., Preconditioned Multigrid Methods for

Unsteady Incompressible Flows. J. Comp. Phys., 139, pp. 35-57, 1998.

[18] Cantwell, B., Coles, D., An experimental study of entrainment and

transport in the turbulent near wake of a circular cylinder. J. Fluid Mech.,

136, pp. 321-374, 1983.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

This page intentionally left blank

www.engbookspdf.com

Advances in Fluid Mechanics IX 109

of foams

J. Skibinski1, T. Wejrzanowski1, J. Szumbarski2 &

K. J. Kurzydlowski1

1

Warsaw University of Technology, Faculty of Materials Science and

Engineering, Poland

2

Warsaw University of Technology, Faculty of Power and Aeronautical

Engineering, Poland

Abstract

In the present studies the quantitative relationships between the structure and

flow properties of porous materials are addressed. In order to investigate such

relationships a series of foam structures with different porosity were

characterized by computer tomography. Later, 3D image analysis was applied to

obtain quantitative parameters of foam structures.

In order to calculate properties of the flow through porous structures, 3D

images obtained by computer tomography were numerically processed and

transformed into a finite element mesh. Simulation of fluid flow was performed

using a Finite Volume Method. Both qualitative characterization and numerical

simulation results, when compared, enabled us to establish structure-flow

relationships.

The preliminary results show that the pressure drop is strongly related to the

porosity of foams. The significant effect on flow properties exhibits also the

diversity of pore sizes.

The results will be discussed with respect to foam structure optimization for

several applications.

Keywords: ceramic foams, pressure drop, finite volume method.

1 Introduction

The industrial application of structures with open porosity such as ceramic or

metallic foams has been growing over recent decade. Their favorable properties

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120101

www.engbookspdf.com

110 Advances in Fluid Mechanics IX

such as high specific surface area, high porosity, low density, low thermal

conductivity and low specific heat capacity predestine them to serve as compact

heat exchangers, reaction catalysts, flow stabilizers or casting filters [1]. These

materials can also be found in nature, for example as porous rocks containing gas

and oil. Those applications require new materials, which can be designed and

investigated numerically before manufacturing. In order to understand

quantitative nature of the structure-properties relationships in porous media

relevant structural parameters must be defined. Open porosity and relative

surface of pores are frequently used for this purpose [2]. These parameters give

good general characteristics, however their meaning to the understanding of the

material flow properties, such as pressure drop, is rather limited.

Many geometrical models have been developed to describe properties of real

foam structures. Most of the models, such as the reproduction of a single type of

polyhedron, or models based on Kelvin polyhedron (gives minimum specific

surface) [3], exhibit anisotropy, which usually does not exist in real foams. Most

of cellular structures are fabricated by techniques based on growth process.

Poisson-Voronoi tessellation (PVT) is frequently used as an algorithm to obtain

representative cellular structures of this type. However, geometrical parameters

of real foams, such as average faces per cell, differ from those obtained by PVT.

A representation of real foams structure can be obtained by application of

Laguerre-Voronoi algorithm, where tessellations are performed on the set of

spheres with pre-determined size distribution [4, 5].

This paper shows the preliminary numerical results of the modelling of fluid

flow properties through real structures and the model structures obtained by

Laguerre-Voronoi tessellation.

2 Geometries

2.1 Real structures

structures (further called real structures). The real structures were originally

described by pores per inch number (ppi). Further 3d image analysis enabled for

extended characterization, where other structural parameters such as porosity,

specific surface, mean pore diameter, mean strut diameter were calculated.

Digitalized geometries of foams were obtained using Computer Tomography

(CT). The vector-like structures were created from set of bitmaps. The surface of

the foam is divided into a logical series of triangles. Each triangle is uniquely

defined by its normal and three vertices. To create a valid volume, the surface

geometry was converted to file of points cloud to simplify model in problematic

areas. As a result, a valid 3D multi-surface mesh was created. The distribution of

the strut diameter as well as specific surface area of each sample were calculated

during post-processing.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 111

Real ceramic foams have varying percentage of pores that are blocked. This is

the effect of fabrication process, and the percentage of clogged pores increases

with number of pores per inch. The resistance to the flow was expected to be

higher in the structures with higher number of blocked pores.

In figure 1 a real geometry and its virtual representation is shown. Table 1

lists structural parameters of foams obtained after processing.

Porosity

Structure diameter

[-] [1/mm]

[mm]

Foam models were created using Laguerre-Voronoi tessellation (LVT). LVT was

applied to spheres with specified size distribution, which were packed in the

cubic volume. The foam struts were created by cylinders with defined constant

diameter generated along cell edges (see Figure 2a).

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

112 Advances in Fluid Mechanics IX

Figure 2: Computer models of designed structure (a) and real structure (b).

[-] [1/mm] [mm]

CV 01 0.6964 0.76741 1.544

CV 03 0.7480 0.75987 1.278

Bimodal 0.8157 0.47446 1.459

The effect of pore clogging does not appear in the model structures. Blocked

pores don’t change porosity significantly, but the pressure drop is expected to be

smaller regardless structural parameters.

3 Numerical method

Fluid flow calculations were performed using Finite Volume Method. This

method represents partial differential equations in the form of algebraic

equations. Values are calculated at discrete place on a meshed geometry. The key

to the method is that the integral form of the conservation law (1) can be

rewritten using the Gauss Divergence Theorem.

Ω

Ω Ω

· Ω 0 (1)

Volume integrals of divergence terms in a partial differential equation are

converted to surface integrals of fluxes all around the control volume (2).

Ω

Ω Ω

· Γ 0 (2)

The resulting equations cannot be solved analytically and a numerical

solution requires discretization into volumes. These terms are then evaluated as

fluxes at the surfaces of each finite volume. The flux entering a given volume is

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 113

identical to that leaving the adjacent volume. This makes the FVM stable and

flexible, yet relatively easy to implement. The main advantage of this method is

it’s applicability on unstructured meshes and the intrinsic local conservation

properties of the resulting schemes, which suits very well the irregular

geometries of porous structures.

The flows across real and designed structures were simulated. The

computational domain was a channel with length of 20 mm and 15x15 mm

square cross-section. Structural parameters of specimens are shown in Table 2.

The size of specimens was 15mmx15mmx15mm block. The inlet velocities for

FVM simulations range from 0.0002 to 0.02 m/s, which correspond to Reynolds

number from 2 to 400.

4 Results

The results of the simulations of flows across real foams are compared with the

designed structures. The hydraulic resistance of all structures is shown in

Figure 3.

across different ceramic foams.

parameters are 10ppi and CV03. These structures have been chosen for

comparison of hydraulic properties of real and designed materials (Figure 4).

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

114 Advances in Fluid Mechanics IX

8

7

6

Pressure drop [Pa]

5

4 CV03

3 10ppi

2

1

0

0 0.005 0.01 0.015 0.02

Velocity [m/s]

structures with similar structural parameters.

0.05

0.045

Pressure drop [Pa]

0.04

0.035 cv01

0.03

cv03

0.025

0.02 Bmodal

0.015 10ppi

0.01 20ppi

0.005

30ppi

0

0.0001 0.0003 0.0005 0.0007

Velocity [m/s]

Linear Darcy’s model is most commonly used for describing flows through

porous media. Darcy’s law may be written as:

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 115

· (3)

where dL is a segment [m] along which a pressure drop dp [Pa] occurs, k

permeability [m2], µ dynamic viscosity [kg/ms] and v is the filtration velocity

[m/s]. Darcy’s law correctly describes the flow in porous media only for low

velocity flows. As a basic criterion used to describe correct Darcy’s regime is

Reynolds number. The range of Reynolds number for Darcy’s flow differs in

dependence of sources [12–15], but Re = 10 is generally assumed to be a critical

maximum value. Set of calculation results based on this criterion have been

chosen to determine effective permeability of materials. Reynolds number valid

for Darcy’s law is equivalent to velocities of 0.0002 m/s to 0,0006 m/s for

porous foams sizes used. Figure 5 shows the results of pressure drop for selected

range of velocities.

For flows with Reynolds number exceeding the Darcy’s law criterion, a

discrepancy between experimental data and results obtained based on Darcy’s

law appears [16]. Forchheimer linked the latter to kinetic effect and suggested to

modify (3) by an additive term ρῡ2, representing kinetic energy [17]:

· (4)

к

coefficient). The results of pressure drop in investigated structures for non-Darcy

flow are shown in Figure 6.

20

18

Pressure drop (Pa)

16

14

12 CV 01

10 CV 03

Bimodal

8

10PPI

6

20PPI

4

30PPI

2

0

0 0.005 0.01 0.015 0.02

Velocity (m/s)

There is a good agreement between obtained results and the basic models

describing flows through porous media. When the Reynolds number exceeds the

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

116 Advances in Fluid Mechanics IX

maximum value for Darcy’s flow, the pressure vs. velocity curve changes from

linear to polynomial for every investigated structure, which accords with

Forchheimer’s equation (4).

The effective permeability keff of materials have been determined from

Darcy’s law. The results are shown in Figure 7.

3E‐06

Permeability [m2]

3E‐06

2E‐06 CV01

CV03

2E‐06

Bimodal

1E‐06

10ppi

5E‐07 20ppi

0E+00 30ppi

0.0001 0.0003 0.0005 0.0007

Velocity [m/s]

characterizing the medium, but also inertial effects. As the contribution of

inertial effects grows with increasing velocity of flow, the effective permeability

decreases.

5 Conclusions

Simulations of flows across porous structures were performed to determine

hydraulic properties. Pressure drop difference between real and designed

structures can be observed. This effect may be related with clogged pores, a

result of the production process. The equivalent model structures considered in

the simulations should be modified to implement blocked pores in the design

process, which should result in more accurate model of real structure.

Acknowledgement

The work has been financially supported by the Polish Ministry of Science and

Higher Education, research grant no. N N507 273636.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 117

References

[1] A. Oscilowski, W. Plis, Ceramic foam filters – using purpose, practical

experience in Ferro-Term, Archives of Foundry, Vol. 1, Book 1, PAN –

Katowice, 2001.

[2] Edouard D., Lacroix M., Huu C.P., Luck F., Pressure drop modeling on

SOLID foam: State-of-the art correlation. Chem. Eng. Journal 144, pp. 299-

311, 2008.

[3] Weaire, D.; Phelan, R. (1994), “A counter-example to Kelvin's conjecture

on minimal surfaces”, Phil. Mag. Lett. 69: 107–110.

[4] Z. Fan, Y. Wu, X. Zhao, Y. Lu, Simulation of polycrystalline structure with

Voronoi diagram in Laguerre geometry based on random closed packing of

spheres, Computational Materials Science 29 (2004) 301–308.

[5] C. Redenbach, Microstructure Models for Cellular Materials,

Computational Materials Science 44, 4 (2009) 1397-1407.

[6] Timothy Bart, Mario Ohlberger, Finite Volume Methods: Foundation and

Analysis, Encyclopedia of Computational Mechanics, Chapter 9, 2004.

[7] M.R. Nangrejo, X. Bao, M.J. Edirisinghe, The structure of ceramic foams

produced using polymeric precursors, Journal of Materials Science Letters

19 (2000) 787-789.

[8] J.G. Fourie, J.P. DuPlessis, Pressure drop modelling in cellular metallic

foams, Chem. Eng. Sci. 57 pp. 2781–2789, 2002. [7] F.C. Buciuman, B.

Kraushaar-Czarnetzki, Ceramic foam monoliths as catalyst carriers. 1.

Adjustment and description of the morphology, Ind. Eng. Chem. Res. 42

pp. 1863–1869, 2003.

[9] V.N. Antsiferov, S.E. Porozova, Foam ceramic filters for molten metals:

reality and prospects, Powder Metallurgy and Metal Ceramics, Vol. 42,

Nos. 9-10, 2003.

[10] Luis Seminario, Roberto Rozas, Rodrigo Borquez, Pedro G. Toledo, Pore

blocking and permeability reduction in cross-flow microfiltration, Journal

of Membrane Science 209 (2002) 121-142.

[11] A. Santos, P. Bedrikovetsky, S. Fontoura, Analytical micro model for size

exclusion: Pore blocking and permeability reduction,. Journal of

Membrane Science 308 (2008) 115-127.

[12] Bear J., Dynamics of Fluids in Porous Media, Elsevier, 1972.

[13] Hansen T. E., Flow in micro porous silicon carbide. Master Thesis,

Department of Micro and Nanotechnology, Technical University of

Denmark, March 2nd, 2007.

[14] Hassanizadeh S.M., Gray W.G., High Velocity Flow in Porous Media,

Transport in Porous Media (1987) 2: 521-531.

[15] Sawicki J., Szpakowski W., Weinerowska K., Woloszyn E., Zima P.,

Laboratory of Fluid Mechanics And Hydraulics, Gdansk University of

Technology, 2004.

[16] Andrade J. S., Costa U. M. S., Almeida M. P., Makse H. A., Stanley H. E.,

Inertial Eﬀects on Fluid Flow through Disordered Porous Media, Physical

Review Letters, Vol. 82, No. 26, 28 June 1999.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

118 Advances in Fluid Mechanics IX

[17] Ewing R., Lazarov R., Lyons S.L., Papavassiliou D.V., Pasciak J., Qin

G.X., Numerical Well Model For Non-Darcy Flow, Computational

Geosciences. 3(3-4). 185-204. 1999.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 119

strategies for local meshless numerical method

G. Kosec1,2, R. Vertnik2, 3 & B. Šarler2,4

1

Jožef Stefan Institute, Laboratory for Parallel and Distributed

Computing, Slovenia

2

University of Nova Gorica, Laboratory for Multiphase Processes,

Slovenia

3

Štore steel, d.o.o. Research, Slovenia

4

Centre of Excellence for Biosensors, Automation and Process Control,

Laboratory for Advanced Materials Systems, Slovenia

Abstract

The performance of two pressure-velocity coupling strategies, associated with

the primitive variable solution of incompressible Newtonian fluid with the

meshless Local Radial Basis Function Collocation Method (LRBFCM), is

compared with respect to computational efficiency, stability, accuracy and

spatial convergence. The LRBFCM is structured with multiquadrics on five

noded support domains. The explicit time stepping is used. The Backward-

Facing Step problem (BFS) has been selected as a benchmark problem,

previously tackled by several numerical methods. The semi-local fractional step

method (FSM) and completely local pressure-velocity couplings (LPVC) are

compared. The numerical results are validated against previously published data.

The results are represented and compared in terms of a convergence plot of the

reattachment position. We show that both approaches provide reasonable results;

however, LVPC is less computationally complex and less stable in comparison

to FSM.

Keywords: meshfree methods, radial basis functions, collocation,

convective-diffusive problems, adaptation, refinement, melting, fluid flow,

Newtonian fluids.

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120111

www.engbookspdf.com

120 Advances in Fluid Mechanics IX

1 Introduction

The meshless methods belong to a class of numerical methods where an

arbitrarily distributed set of nodes, without any additional topological relations

between them, is used. There exist several meshless methods [1–5]; however,

this work is focused on one of the simplest classes of meshless methods in

development today, the point interpolation Radial Basis Function Collocation

Method (RBFCM) [6]. In the present paper we use a local variant of RBFCM

[6], the Local Radial Basis Function Collocation Method (LRBFCM). The main

advantage of the local approach is in consideration of the multivariate data fitting

and calculation of spatial derivatives through a local support domain.

Consequently, the computational basis is simplified, since several small systems

of algebraic equations are solved instead of a global algebraic equation system.

Such approach has been already successfully applied to several thermo-fluid

situations, ranging from the basic diffusion problems to complex highly non-

linear and coupled technologically relevant situations [5, 7–14]. Important

features of meshless methods represent straightforward application of

dynamic node distribution [15], and stable behaviour on non-regular node

distributions [11].

This paper is focused on application of LRBFCM in fluid mechanics, more

precisely, in channel flow. In our past publications we have already

demonstrated the applicability of LRBFCM in fluid flow computations. The

LPVC (in the context of meshless methods) was for the first time introduced in

[10] and further investigated regarding more complex physical systems in

[12, 16, 17], where only closed and impermeable domains have been considered.

The open domain problems were dealt with (in the context of meshless methods)

the FSM approach in [14, 18].

In this paper we compare the performance of LPVC and FSM in open domain

problems, which seems to appear more demanding than the closed ones. The

well-known Backward-Facing Step problem [19–25] (BFS) is considered since

its solution is well coped with in the literature, and is at the same time complex

enough to challenge the proposed numerical techniques.

The principal message of the present paper is a quantitative comparison of

two meshless based numerical procedures that differ only in pressure-velocity

coupling strategies. The Chorin’s FSM [21] is semi-global, and as such requires

solution of a global system of algebraic equations for pressure, while LPVC is

completely local. Both strategies show reasonably good convergence behavior as

well as good agreement with already published data [22].

2 Problem definition

The BFS problem of the laminar incompressible Newtonian fluid flow over a

backward-facing step in two dimensions is considered. It represents a standard

test for investigating the flow separation and reattachment. It has been already

used by numerous researchers as a benchmark test for various numerical

methods [19–25]. The domain of the problem is characterized by a planar

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 121

channel with sudden change of the geometry (Figure 1), which governs the flow

separation, and generation of several re-circulating zones downward the step,

described by positions px1 , px 2 , px3 . The flow is modelled by the Navier

Stokes equations

v 0 , (1)

v

( vv) P v , (2)

t

with t , v v x , v y , P , and standing for time, velocity, pressure, viscosity and

density, respectively. The Cartesian coordinates are used, and the position vector

is defined as p px , p y . The rectangular domain is normalized to height

2h 1 . Beginning of the computational domain is considered at the location of

the step, where the flow enters the channel. The step height is set to h . The

channel length is set to 30h . The flow dynamics is characterized by the

Reynolds number (Re), defined as

v0 2h

Re , (3)

with v0 standing for the amplitude of the inlet velocity. At the inlet, the fully-

developed velocity profile is set [22]

components are prescribed and are all set equal to zero. At the walls, non-

permeable and no-slip conditions are used. The schematics of the computational

test are represented in Figure 1.

¶P

=0

px 2 ¶y px 3

¶vx

vx = vin =0

inlet ¶x

¶P ¶v y

=0 2h outlet

¶x =0

¶x

px1

P=0

¶P

=0 v=0

¶y

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

122 Advances in Fluid Mechanics IX

3 Solution procedure

A general idea in the local meshless methods is the use of local sub clusters of

domain nodes, named local support domains, for the approximation of the fields.

Complementary to the selected support domain, an approximation function is

introduced as a sum of weighted basis functions

N Basis

(p)

n 1

n n (p), (5)

where , N Basis , n and n stand for the approximation function, the number of

basis functions, the approximation coefficients, and the basis functions,

respectively. The basis could be selected arbitrarily (monomials, radial basis

functions, …). In this paper, Hardy’s Multiquadrics (MQs),

n p p p p p /

n n 2

C 1, (6)

with C standing for the free shape parameter of the basis function, are used.

MQ’s are selected. By taking into account all support domain nodes and equation

(5) the approximation system is obtained. In this paper we use collocation (the

number of support nodes is the same as the number of the basis functions). An

arbitrary spatial differential operation L can be applied on the approximation

function in the following way

N Basis

L p

n 1

n L n (p) (7)

In general, the system (5) has to be solved only when the support domain

topology changes and therefore the computation can be optimized by computing

Ψ1 in a pre-process. Furthermore, the computation of the coefficients and the

evaluation of differential operators can be combined. All information about the

numerical approach and the local nodal topology can be stored in a predefined

vector, which has to be re-evaluated only when the topology of the nodes

changes. The differential operator vector mL is introduced as

N

mL (p) nm

1

L n (p) . (8)

n 1

The introduced formalism holds in general and therefore the general notation

for partial differential operator L is used. However, in the present work, only

operators and 2 are employed

N

2

m (p) nm1 n (p),

2

(9)

n 1 p2

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 123

N

m / p (p) nm

1

n (p). (10)

n 1 p

The structured formulation is convenient since most of the complex and CPU

demanding operations are performed in the pre-process phase. For all inner

temporal loop operations only N floating point operations (FLOPS) are needed

for evaluation of an arbitrary partial differential operator. The implementation of

the Dirichlet boundary condition is straightforward. In order to implement

Neumann and Robin boundary conditions, special interpolation is needed. In the

present numerical framework, the computation of Neumann and Robin boundary

conditions can be simplified through the use of the differential operator vector.

Consider Neumann boundary condition

a b BC , (11)

n

N Sub

BC a m / n m

0 m2

, (12)

a 0 / n b

where 0 stands for the boundary node. Equation (12) simplifies to Neumann

boundary condition computation if b is set to zero. Such approach makes the

Neumann and the Robin boundary condition computation straightforward and

CPU effective. Again, only N flops are needed to evaluate it, without any kind

of special computational treatment on or near the boundaries.

t

vˆ v 0

P v b

0 0 0 (v0 v0 ) , (13)

with t standing for time step, index zero marks the initial time step values and

v̂ stands for intermediate velocity. Equation (13) does not take into account the

mass continuity and therefore the pressure and the velocity corrections are added

vˆ m 1 vˆ m v , Pˆ m 1 Pˆ m P , (14)

where m, v and P stand for pressure velocity iteration index, velocity

correction, and pressure correction, respectively. By combining the momentum and

the mass continuity equations, the pressure correction Poisson equation emerges

t

vˆ m 2 P. (15)

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

124 Advances in Fluid Mechanics IX

correction is directly related to the intermediate velocity divergence

P 2 vˆ m . (16)

t

The proposed assumption enables direct solving of the pressure velocity

coupling iteration. It is thus very fast due to only one spatial discretization

operation needed in each node to evaluate the new iteration pressure and the

velocity corrections. With the computed pressure correction, the pressure and the

velocity can be corrected as

t

vˆ m 1 vˆ m P, Pˆ m 1 Pˆ m P, (17)

where stands for the relaxation parameter. The iteration is performed until the

criterion ·vˆ V is met in all computational nodes. After successful pressure-

velocity iteration, the algorithm continues with the next time step. The LPVC

approach is similar to the artificial compressibility method (ACM), which has

been recently under intense research in connection with Finite Volume Method

[26, 27] and in connection with Finite Element Method [28]. A similar approach,

in the framework of the FDM, is SOLA algorithm [29]. However, the proposed

LPVC approach retains the correct time transient which is, in general, not the

case in SOLA and ACM approaches.

in FSM approach

t

vˆ v 0

v b

0 0 ( v 0 v 0 ) . (18)

Again, the equation (18) does not take into account the mass continuity. In a

similar manner as in 3.1, the pressure Poisson equation is constructed.

t

2 P vˆ . (19)

The equation (19) is directly solved in the FSM approach. However, the

locality of spatial discretization simplifies the computations to solution of a

sparse system of algebraic equations. With the solved pressure, the intermediate

velocity is corrected as

t

v vˆ P . (20)

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 125

In FSM, the internal iterations are not performed. The computed pressure and

corrected velocity are therefore new values and simulation proceeds to the next

time step. FSM does not provide a correct time transient in general.

4 Results

The results are presented in terms of velocity field plot and velocity, pressure

and shear stress cross section profiles for Re 800 case. The convergence is

monitored at the reattachment position px1 with respect to the number of

computational nodes, ranging from 150x40 to 325x110. The reattachment

position is defined as a point where the wall shear stress equals zero

vx

0 (21)

p y

Steady state velocity field is represented in Figure 2. The x axis is on the plot

limited to 10 (for the sake of better visibility), although the computations are

performed on domain with length 15.

0.5

py

0

0 1 2 3 4 5 6 7 8 9 10

px

parameter 10 7 and MQ shape parameter C 32 , normalized to the

maximum distance between the nodes within the support domain.

A comparison of the results is done on velocity cross-section profiles with the

already published data [22], presented as “Gartling” in Figures. Two different

cross-sections are examined, at px 7 and px 15 (Figure 3). In Figure 4, the

shear stress profiles at the bottom and the top walls are represented. A good

agreement is achieved in all analyses. Finally, the convergence plot for both

meshless based solutions is represented in Figure 5. Both solution procedures

converge at the same rate with minimal differences. However, it can be seen that

LPVC develops minor instabilities in the convergence plot. The behaviour is to

be investigated in our future work. Our preliminary analyses show that the

instabilities can be mitigated by usage of an additional stability term in pressure

correction.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

126 Advances in Fluid Mechanics IX

px = 7 px = 15

1.2 1.2

LPVC LPVC

FSM FSM

1 Garltling 1 Garltling

0.8 0.8

0.6 0.6

vx

vx

0.4 0.4

0.2 0.2

0 0

0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

py py

−3

0.01 x 10

5

0

0.005

τ (py = 0)

τ (py = 1)

−5

0

−10

−0.005

LPVC −15 LPVC

FSM FSM

Garltling Garltling

−0.01 −20

0 5 10 15 0 5 10 15

px px

Figure 4: Comparison of shear stress profiles at the top and at the bottom wall.

FSM

0.702 LPVC

0.7

0.698

log10 (px1 )

0.696

0.694

0.692

0.69

0.688

0.686

3.8 4 4.2 4.4 4.6

log10 (ND )

Figure 5: Convergence plot for reattachment position px1 with respect to the

number of the nodes ND .

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 127

5 Conclusions

Two different strategies for pressure-velocity coupling in meshless solution of

laminar, incompressible Newtonian channel flow are presented. Both strategies

provide good results. Comparison with published data shows good accuracy of

both presented methods. The semi-local FSM behaves more stable as the LPVC.

The main advantages of the LPVC approach are: ease of implementation,

straightforward parallelization, and low computational cost. On the other hand,

the global FSM pressure-velocity strategy does not require a consideration of any

extra relaxation parameter, at the cost of higher computational complexity. The

FSM approach also shows more stable convergence behavior. The main

drawback of FSM as compared to the LPVC is global solution of pressure

equation. First, the global system has to be constructed. This is not problematic

as long as the static node distribution is used, as it can be done in preprocess.

The solution of the system remains the task to be done in each time step,

however. With proper iterative algorithms and pre-conditioning this overhead

can be minimized. Still, the complexity is proportional to O ND , while for

LPVC this step is omitted. Besides a higher computational complexity, the

important factor is ease of parallelization since the multicore computers are

common today. The LPVC is almost ideally parallelizable as it is completely

local. There is minimal communication with other parts of the computational

domain. For the shared memory systems the OpenMP parallelization is trivial

and extremely effective [30]. At the cost of computational complexity, however,

the FSM offers more stable numerical scheme. The FSM also does not provide a

proper time transient. These conclusions are preliminary and will be further

investigated and supported with more analyses in the archival version of the

present paper, that will be submitted to EABE.

Acknowledgements

The research was funded through Slovenian state research projects J2-4120,

P2-0379 and P2-0095. The Centre of Excellence for Biosensors, Instrumentation

and Process Control is an operation financed by the European Union, European

Regional Development Fund and Republic of Slovenia, Ministry of Higher

Education, Science and Technology. The financial support is kindly

acknowledged.

References

[1] Kansa, E. J., Multiquadrics - a scattered data approximation scheme with

application to computational fluid dynamics, part I. Computers and

Mathematics with Applications, 19, pp. 127-145, 1990.

[2] Chen, W., New RBF collocation schemes and kernel RBFs with

applications. Lecture Notes in Computational Science and Engineering, 26,

pp. 75-86, 2002.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

128 Advances in Fluid Mechanics IX

[3] Liu, G.R., Mesh Free Methods, CRC Press: Boca Raton, 2003.

[4] Trobec, R., Šterk, M. and Robič, B., Computational complexity and

parallelization of the meshless local Petrov-Galerkin method. Computers &

Structures, 87, pp. 81-90, 2009.

[5] Šterk, M. and Trobec, R., Meshless solution of a diffusion equation with

parameter optimization and error analysis. Engineering Analysis with

Boundary Elements, 32, pp. 567-577, 2008.

[6] Šarler, B., From global to local radial basis function collocation method for

transport phenomena. Advances in Meshfree Techniques, pp. 257-282, 2007.

[7] Šarler, B. and Vertnik, R., Meshfree explicit local radial basis function

collocation method for diffusion problems. Computers and Mathematics

with Applications, 51, pp. 1269-1282, 2006.

[8] Vertnik, R. and Šarler, B., Meshless local radial basis function collocation

method for convective-diffusive solid-liquid phase change problems.

International Journal of Numerical Methods for Heat and Fluid Flow, 16,

pp. 617-640, 2006.

[9] Divo, E. and Kassab, A. J., Localized meshless modeling of natural-

convective viscous flows. Numerical Heat Transfer, B129, pp. 486-509,

2007.

[10] Kosec, G. and Šarler, B., Solution of thermo-fluid problems by collocation

with local pressure correction. International Journal of Numerical Methods

for Heat and Fluid Flow, 18, pp. 868-882, 2008.

[11] Trobec, R., Kosec, G., Šterk, M. and Šarler, B., Comparison of local weak

and strong form meshless methods for 2-D diffusion equation. Engineering

Analysis with Boundary Elements, 36, pp. 310-321, 2012.

[12] Kosec, G., Založnik, M., Šarler, B. and Combeau, H., A Meshless

Approach Towards Solution of Macrosegregation Phenomena. Computers,

materials & continua, 580, pp. 1-27, 2011.

[13] Vertnik, R., Založnik, M. and Šarler, B., Solution of transient direct-chill

aluminium billet casting problem with simultaneous material and interphase

moving boundaries by a meshless method. Engineering Analysis with

Boundary Elements, 30, pp. 847-855, 2006.

[14] Vertnik, R. and Šarler, B., Local collocation approach for solving turbulent

combined forced and natural convection problems. Advances in Applied

Mathematics and Mechanics, 3, pp. 259-279, 2011.

[15] Kosec, G. and Šarler, B., H-Adaptive Local Radial Basis Function

Collocation Meshless Method. CMC: Computers, Materials & Continua,

26, pp. 227–254, 2011.

[16] Kosec, G. and Šarler, B., Local RBF collocation method for Darcy flow.

CMES: Computer Modeling in Engineering & Sciences, 25, pp. 197-208,

2008.

[17] Kosec, G. and Šarler, B., Solution of phase change problems by collocation

with local pressure correction. CMES: Computer Modeling in Engineering

& Sciences, 47, pp. 191-216, 2009.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 129

mesh-free method. CMES: Computer Modeling in Engineering & Sciences,

44, pp. 66-95, 2009.

[19] Armaly, B. F., Durst, F., Pereira, J. C. F. and Schönung, B., Experimental

and theoretical investigation of backward-facing step flow. Journal of Fluid

Mechanics, 127, pp. 473–496, 1983.

[20] Chiang, T. P. and Sheu, T. W. H., A numerical revisit of backward-facing

step flow problem. Physics of fluids, 11, pp. 862, 1999.

[21] Chorin, A.J., A numerical method for solving incompressible viscous

flow problems. Journal of Computational Physics, 2, pp. 12-26, 1967.

[22] Gartling, D. K., A test problem for outflow boundary conditions—flow

over a backward-facing step. International Journal for Numerical Methods

in Fluids, 11, pp. 953–967, 1990.

[23] Gresho, P. M., Gartling, D. K., Torczynski, J. R., Cliffe, K. A., Winters, K.

H., Garratt, T. J., Spence, A. and Goodrich, J. W., Is the steady viscous

incompressible two-dimensional flow over a backward-facing step at Re=

800 stable? International Journal for Numerical Methods in Fluids, 17,

pp. 501–541, 1993.

[24] Ramšak, M. and Škerget, L., A subdomain boundary element method for

high-Reynolds laminar flow using stream function-vorticity formulation.

International Journal for Numerical Methods in Fluids, 46, pp. 815–847,

2004.

[25] Zhang, X. H, Ouyang, J. and Zhang, L., Characteristic based split (CBS)

meshfree method modeling for viscoelastic flow. Engineering Analysis with

Boundary Elements, 34, pp. 163–172, 2010.

[26] Rahman, M. M. and Siikonen, T., An artificial compressibility method for

viscous incompressible and low Mach number flows. Internation Journal of

Numerical Methods in Engineering, 75, pp. 1320-1340, 2008.

[27] Malan, A. G. and Lewis, R. W., An artificial compressibility CBS method

for modelling heat transfer and fluid flow in heterogeneous porous

materials. International Journal for Numerical Methods in Engineering, 87,

pp. 412-423, 2011.

[28] Traivivatana, S., Boonmarlert, P., Thee, P., Phongthanapanich, S. and

Dechaumphai, P., Combined adaptive meshing technique and

characteristic-based split algorithm for viscous incompressible flow

analysis. Applied Mathematics and Mechanics, 28, pp. 1163-1172, 2007.

[29] Hong, C. P., Computer Modelling of Heat and Fluid Flow Materials

Processing, Institute of Physics Publishing Bristol, 2004.

[30] Kosec, G., Trobec, R., Depolli, M. and Rashkovska, A.: Multicore

parallelization of a meshless PDE solver with OpenMP, ParNum 11, 2011.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

This page intentionally left blank

www.engbookspdf.com

Advances in Fluid Mechanics IX 131

supersonic nozzle

E. Eslamian, H. Shirvani & A. Shirvani

Faculty of Science and Technology, Anglia Ruskin University, UK

Abstract

This study reports the Computational Fluid Dynamics (CFD) results of a swirling

flow induced by introducing a helical insert inside a supersonic nozzle. The CFD

simulation shows a very complex unsteady, non-axisymmetric flow pattern for

the swirl flow inside the nozzle. The flow is investigated by solving the

Reynolds averaged Navier Stokes (RANS) equations with k- and Reynolds

Stress Model (RSM) turbulence models to predict the flow patterns and the type

of swirling flow. Computations are conducted for a range of nozzle pressure

ratios with and without swirl inside the converging–diverging nozzle. The study

has revealed a new understanding and data for flow features such as shock

location, mass flow rate and anisotropic turbulence.

Keywords: swirl flow, supersonic nozzle, flow separation, CFD.

1 Introduction

Investigation of supersonic flow inside converging–diverging (C–D) nozzles has

been the subject of several numerical and experimental studies in the past [1, 2]

but there is not that much research on the effect of swirl flow inside C–D

nozzles. Swirling flows which are very common in technical applications, such

as turbo machinery, cyclones or separators, and they require sophisticated

modeling.

The effect of swirl inside a nozzle can improve the mixing features of the

flow by increasing turbulence and vorticity in the nozzle, which can be useful in

combustion injectors and sand blast techniques; also swirl flow will change the

shock structure and its interaction with the boundary layer and create a larger

separation zone at the exit of a nozzle. These viscous and compressible

phenomena affect the flow behavior inside and outside of a nozzle. Since

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120121

www.engbookspdf.com

132 Advances in Fluid Mechanics IX

experimental data for the swirl flow inside a high speed nozzle are scare,

numerical results are vital for understanding the flow and for further analysis.

In this study the finite volume method is utilized (Fluent), to solve the

governing equations of flow.

2 Numerical analysis

2.1 Nozzle geometry

The dimension of the nozzle with the flow inside which has been investigated is

shown in Figure 1a. The length of the nozzle is 200mm and it has three sections.

The first 64mm is the convergence section, then comes a 16mm section with

constant 11mm diameter and this is followed by the divergence section with

length 120mm and outlet diameter of 15mm. The computational domain which

consists of the helical inlet and the nozzle is illustrated in Figure 1b. The helical

section, Figure 1c, is used to create swirl flow inside the nozzle. The helical

section has a 31.75mm diameter with length of 76.45mm; the spiral part has two

revolutions with a start angle of 45 degrees.

(a)

(b) (c)

Dynamics (CFD). The models mostly used for fluid phenomena are the standard

k- and k- models. These Reynolds averaged Navier Stokes (RANS) models

give adequate results for isotropic turbulence flows, and have been validated

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 133

with many experimental tests (Wilcox [3]). However, as we have highly swirling

flow inside nozzle, there will be a considerable degree of anisotropy in the stress

and dissipation tensors which causes a highly anisotropic eddy viscosity (Yajnik

and Subbaiah [4]), so the isotropic turbulence model will not give precise results.

To be able to capture anisotropic turbulence inside a nozzle we have to consider

all Reynolds stress equations. The highest accuracy in turbulence is acheived by

Direct Numerical Simulations (DNS), but this needs huge processing power.

Large Eddy Simulations (LES) is another major turbulence model which lately

has begun to grow in popularity and many academic and industry users deploy it

for their research and applications. LES resolves all anisotropic turbulent

structures both in time and space, and just leaves the small scales to be modelled

by simple turbulence models (Ogor et al. [5]). However, we still need a lot of

precessing power and a very fine mesh to use LES modelling.

The RANS models for anisotropic turbulence are limited to the modified k-

and Reynolds Stress Model (RSM). We have used a realizable k- model, which

contains an alternative formulation for eddy viscosity and uses a modified

transport equation for dissipation rate (Shih et al. [6]) but the RSM model solves

all six transport equations. In this research, the RSM gave more consistent

results, and for a realizable k- method we need to carry out more research with a

higher order scheme; therefore the focus is more on the RSM and the results

shown here are based on the RSM.

For the RSM a second oreder upwind scheme is applied to the flow equations,

but the turbulent kinetic energy, turbulent dissipation rate and Reynolds Stress

are considered first order upwind.

The initial simulations were conducted with a steady method, but the

convergence history of residuals showed unsteadiness which is in agree with the

results of Xiao et al. [7]; the main reason for this is the separation of the flow at

the exit of the nozzle. Therefore a transient solver is considered for the flow

with 1e-5 time step intervals.

The air in the nozzle will reach 1, so compressibility is one of the main

factors for the flow solver. The density based solver in Fluent gives more

accurate results and has better convergence rate for compressible flows. It also

captures more precicely the shock waves in the nozzle .

The density based solver from Fluent has been used to solve the governing

equations of the flow. The conservative integral form of the continuity,

momentum and energy equations for a single component fluid in an infinitesimal

control volume is (Fluent [8]):

. (1)

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

134 Advances in Fluid Mechanics IX

0

̂

, ̂ , (2)

The vector H contains all of the source terms such as the body forces. In the

above equation ρ is density, is velocity, is total energy per unit mass and is

pressure. and q are the viscous stress tensor and heat flux, respectively. In the

density based solver of Ansys Fluent, to overcome the poor convergence rate of

high speed flows, a preconditioning technique has been deployed. The

preconditioning technique modifies the time derivative term of equation (1) by

multiplying it with a preconditioning matrix. This will rescale the acoustic speed

of the governing equations in order to mitigate the numerical stiffness

encountered in low Mach numbers.

In Reynolds averaging the variables of exact N-S equations are substituted with

mean and fluctuation components. By substituting into the

momentum equation, where is the mean and is the fluctuating velocity

components, the time average of the momentum equation can be shown as:

(3)

In order to close equation (3), the Reynolds stress term ( ) must be

modelled. In this study we have used the RSM, which uses the exact transport

equation for the transport of the Reynolds stresses, , which can be

written as:

, , (4)

molecular diffusion term, equals the stress production, is buoyancy

production, stands for pressure strain, is the dissipation and is

production by the system rotation. Four parameters, , , , and , have

to be modelled in order to close equation (4), but the rest of the parameters do

not required modelling. The turbulent diffusion ( , ) has been modelled with

the simplified model of Daly and Harlow [9]:

, (5)

where is the turbulent viscosity. The pressure strain term has been modelled

according to the proposal by Gibson and Launder [10]. Buoyancy production

( ) for an ideal gas is modelled as:

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 135

(6)

, 2 (7)

model by Sarkar and Balakrishnan [11].

The results for two pressure ratios ⁄ of 2 and 3 are shown here. At both

pressure ratios there is an overexpanded supersonic nozzle, where the shock

waves occur inside the nozzle. The Mach number contours for a pressure ratio of

2 in two time steps are illustrated in Figure 2. Experimental and numerical

simulations represent two separation patterns for an overexpanded nozzle, the

Free Shock Separation (FSS) and Restricted Shock Separation (RSS) (Hadjadj

and Onofri [12]). In FSS the separation region extends from the separation point

to the end of the nozzle, but in RSS the separation zone reattaches to the surface

and creates recirculation bubbles. For a pressure ratio of 2 the FSS pattern can be

seen at the upper side of the nozzle wall (Figure 2).

(a)

(b)

Figure 2: Mach number contours for a pressure ratio of 2 at (a) t=0.8e-2s and

(b) t=1.8e-2s.

Because of asymmetry which is due to the Coanda effect, flow tends to attach

to the surface and creates a RSS pattern on the down side of the nozzle wall. The

separation zone starts at the interaction point between the oblique shock wave

and the nozzle wall which creates an adverse pressure gradient. Reflection of the

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

136 Advances in Fluid Mechanics IX

oblique shock wave on the shear layer generates expansion fans. To provide

more detail of the separation nozzle, Figure 3 illustrates the flow and shock wave

pattern for the pressure ratio of 2. Since the separation zone is large the flow

remains attached to the down side of the nozzle wall, and therefore the difference

between two time steps is mostly about the change in the shear layer position.

Separation zone

Oblique

shock

Shear layer

Expansion Mach

Mach fan Disk 2

Disk 1

Nozzle exit

Recirculation bubble

Figure 3: Schematic of the shock wave structure for the pressure ratio of 2 at

the exit of the nozzle.

By increasing the pressure ratio to 3 the oblique shock waves become weaker

so the pressure gradient behind it will significantly reduce and the separation

point moves downstream of the nozzle. Contours of Mach number for two time

steps at the exit of the nozzle are shown in Figure 4.

1.9

1.8

1.7

(a) 1.6

1.5

1.4

1.3

1.2

1.1

1

0.9

0.8

0.7

0.6

(b) 0.5

0.4

0.3

0.2

0.1

Figure 4: Mach number contours for the pressure ratio of 3 at (a) t=o.8e-2s,

(b) t=2.8e-2s.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 137

The flow pattern still remains asymmetric, and the second Mach disk moves

out of the nozzle. At t=0.8e-2, Figure 4a, the second Mach disk is at the exit of

the nozzle but in t=2.8e-2, Figure 4b, it is completely out of the nozzle.

Calculation of 1000 more time steps does not show any major change to the

shock wave structure apart from the shear layer movement. The reflection of the

shock wave on the shear layer creates expansion fans.

By adding swirl to the flow the shock wave structure becomes weaker, and

instead of one strong shock wave there will be series of shock waves and

expansion fans at the exit of the nozzle. This can be seen from the static pressure

diagram at the centre of the nozzle which for a pressure ratio of 2 is shown in

Figure 5. Also the start point of the shock wave has been moved upstream for the

swirl flow. The results are compared with experimental data from Abbasalizadeh

[13] for the same nozzle without the swirl that shows good agreement with CFD

calculations.

30

25

20

15

P (psi)

10

5

0

‐0.08 ‐0.03‐5 0.02 0.07 0.12 0.17 0.22

‐10

x (m)

Figure 5: Static pressure diagram for pressure ratio 2 at the centre of nozzle.

can see that for the nozzle without swirl there is no shock wave inside the nozzle,

but by adding swirl conditions, the nozzle remains overexpanded; however, the

shock waves become weaker compared to those for a pressure ratio of 2. The

experimental data from Abbasalizadeh [13] for the nozzle without swirl also

show that there is no shock wave inside the nozzle and the nozzle is in an

underexpanded condition.

Apart from the exit of the nozzle for both pressure ratios the rest of the centre

line static pressure of the nozzle with swirl matches that of the nozzle without

swirl.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

138 Advances in Fluid Mechanics IX

t=2.8e‐2 t=0.8e‐2

No swirl Experiment‐no swirl

45

35

25

P (psi)

15

‐15

X (m)

Figure 6: Static pressure diagram for the pressure ratio of 3 at the centre of the

nozzle.

Since it is very difficult to carry out experimental tests inside a nozzle and there

are little experimental data for supersonic swirl flow inside a nozzle, the

validation process is very important to gain confidence in the results. To check

grid dependency all the simulations have been run for 200k, 450k and 800k cells

with refinement at the boundaries and exit of the nozzle to capture the boundary

layer and shock waves. We obtain similar results for three different meshes.

Considering the computational effort, the results shown here are based on a 400k

mesh. The residuals are all reduced at least by an order of 4. Because of the 3D

complex geometry, it was difficult to get a high quality hexahedral mesh;

therefore to obtain a good quality mesh a tetrahedral mesh with a prism layer at

the boundaries has been used. The minimum quality for the mesh was 0.84. In

addition, the mass flow rate through the inlet and outlet is checked to ensure that

conservation of mass is observed. The net mass flow rate through the system is

under 1%.

The maximum wall yplus value for the CFD tests was 50, and the RSM which

is a core-turbulent model gives better results with a yplus value in the log-low

region (y+>30 to 60) (Salim and Cheah [14]), so the yplus value is satisfied for

most of the nozzle wall.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 139

3 Conclusion

The CFD analysis of swirling flow inside a converging–diverging nozzle is

studied at pressure ratios of 2 and 3. For turbulence modelling, the RSM has

been used to solve seven Reynolds stress equations. The following conclusions

can be drawn:

1. Numerical solutions show swirling flow has unsteady asymmetric

behaviour, with a separation zone at the exit of the nozzle, and the main

reason for asymmetry is the Coanda effect.

2. Because of anisotropic turbulence, a two equation model will not solve all

the scales of turbulent flow, so the RSM with a second order scheme is

selected for turbulence modelling.

3. By increasing the pressure ratio to 3 the shock waves move out of the

nozzle in no swirl conditions, but by adding swirl condition to the flow

the nozzle design criteria will be changed and the nozzle still remains

overexpanded.

4. The centre line static pressure shows in swirl conditions that, instead of

having one strong shock wave, there are a series of weak shock waves

and expansion fans.

References

[1] Anderson, J. D., Modern compressible flow with historical perspective,

New York: McGraw-Hil, 2003.

[2] Anderson, J. D., Hypersonic and high temperature gas dynamics, American

Institute of Aeronautics and Astronautics, 2000.

[3] Wilcox, D. C., Turbulence Modelling for CFD, California: Griffin, 1994.

[4] Yajnik, K. and Subbaiah, M., “Experiments on swirling turbulent flows.

Part 1. Similarity in swirling flows,” Journal of Fluid Mech, vol. 60, no. 4,

pp. 665-687, 1973.

[5] Ogor, I. B., Gyllenram, W., Ohlberg, E., Nilsson, H. and Ruprecht, A., “An

Adaptive Turbulence Model for Swirling Flow,” in Conference on

Turbulence and Interactions, Porquerolles, France, 2006.

[6] Shih, T., Liou, W., Shabbir, A., Yang, Z. and Zhu., J., “A New k -e Eddy-

Viscosity Model for High Reynolds Number Turbulent Flows - Model

Development and Validation,” Computers Fluids, vol. 24, no. 3, pp. 227-

238, 1995.

[7] Xiao, Q., Tsai, H. and Papamoschou, D., “Numerical Investigation of

Supersonic Nozzle Flow Seperation,” AIAA, pp. 532-541, 2007.

[8] Ansys, “Ansys Fluent 13 User's Guide,” Ansys, Inc, 2010.

[9] Daly, B. and Harlow, F. H., “Transport Equations in Turbulence,” Physics

of Fluids, vol. 13, no. 11, p. 2634–2649, 1970.

[10] Gibson, M. and Launder, B., “Ground Effects on Pressure Fluctuations in

the Atmospheric Boundary Layer,” Journal of Fluid Mechanics, vol. 86,

no. 3, p. 491–511, 1978.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

140 Advances in Fluid Mechanics IX

Turbulence Model to the Compressible Shear Layer,” NASA Langley

Research Center, Virginia, 1990.

[12] Hadjadj, A. and Onofri, M., “Nozzle Flow Separation,” Shock Waves, pp.

163-169, 2009.

[13] Abbasalizadeh, M., Investigation of Three Phase Nozzle flow In an

Innovative Sand Blasting System, Chelmsford: Anglia Ruskin University,

2011.

[14] Salim, M. S. and Cheah, S., “Wall yplus Strategy for Dealing with Wall-

bounded Turbulent Flows,” in Proceeding of the International

MultiConference of Eng. and Comp., Hong Kong, 2009.

[15] Guo, H.-F., Chen, Z.-Y. and Yu, C.-W., “3D numerical simulation of

compressible swirling flow induced by means of tangential inlets,”

International Journal For Numerical methods in Fluids, pp. 1285-1298,

2008.

[16] Najafi, A., Saidi, M., Sadeghipour, M. and Souhar, M., “Numerical

analysis of turbulent swirling decay pipe flow,” International

Communications in Heat and Mass Transfer, pp. 627-638, 2005.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 141

two-dimensional wings

S. Djellal, T. Azzam, M. Djellab & K. Lakkaichi

Fluid Mechanics Laboratory Polytechnical School Bordj El Bahri,

Alger, Algeria

Abstract

The flow around a NACA 641-412 airfoil with circle damage is numerically

investigated using the Ansys Fluent package. Several diameter values are

considered for the damage which is located at either quarter or mid chord. The

numerical domain is covered with a multiblock structured hexagonal grid

consisting of 344,680 cells in the undamaged case and 351,540 cells in the

damaged case. Inside the damage hole, a structured tetragonal mesh is used.

Turbulence effects are taken into account via the k- model. The results show

that the presence of the damage hole decreases the lift coefficient and increases

the drag coefficient, resulting in a loss of airfoil performance (fineness decrease).

The numerical simulations show that the flow through the damage corresponds

either to a weak or a strong jet. In the first case an attached wake forms giving

the smallest change in the force coefficients whereas the second case shows a

separated wake with a reverse flow giving the highest force coefficient change.

The present paper also compares the structure of the damage through flow with

previously published experimental results. Finally, the numerical solution is

qualitatively and quantitatively validated using available experimental results.

Keywords: damage, drag, lift, strong jet, weak jet, undamaged.

1 Introduction

The survivability of an aircraft is becoming one of the key aircraft design

requirements [1] and is dependant upon its vulnerability to damage caused by a

variety of threats. Published works to date in this field focus mainly on two-

dimensional wings.

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120131

www.engbookspdf.com

142 Advances in Fluid Mechanics IX

[2, 3] and by many other authors; Render et al. [4], Mani and Render [5] and

Djellal and Ouibrahim [6]. Recently Saaedi et al. [7] published a numerical study

in this field.

In this investigation, we are interested in the aerodynamic performance losses

caused to a two dimensional wing by the impact of battle damage. For this

matter, we investigate by means of numerical methods the undamaged case and

damaged case by considering the variation of the hole diameter and its chordwise

position.

In this work, we analyse the flow field inside the damage hole and also

around the undamaged and damaged aerofoil. In order to validate the numerical

results, obtained values have been compared with those of experimental

investigations led by the previous authors [2, 5, 6, 8].

2 Damage modelling

The most common type of damage used in simulations is the circular hole [6].

The study of other shapes reported by Mani and Render [5] has not shown

noticeable differences.

Damage size can be expressed in terms of a percentage diameter d to the local

chord length c. The range of damage sizes varies from 0.1 to 0.4 c [2] and, for

our simulations, two diameters, 0.2, 0.3, have been considered and both were

located at mid span.

NACA 641-412 aerofoil experimented firstly by Render and his colleagues

serves as the model’s cross section. The aerofoil’s chord and the wing’s span are

200 mm and 450 mm respectively (Fig. 1).

We only considered wing damaged at the quarter and the mid-chord, because

they produced the greatest adverse influence on aerodynamic performance [2].

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 143

The numerical domain is made discrete by two approaches: (1) structured

tetragonal cells for circle damage (2) multiblock structured hexagonal cells are

applied for the remainder damaged aerofoil. The finite volume discretisation

method is used in ‘Fluent’ numerical code.

Following a suitable study of grid sensitivity, Number of cells applied in this

numerical domain is 344,680 cells in the undamaged case and 351,540 cells in

the damaged case. It is worth mentioning that the used structured mesh that is

able to capture the boundary layer was a complicated and time consuming

process. The multiblock approach generated in Gambit allows via the interface

condition to reduce the cells number and consequently the time process.

The results did not show a major difference between aerodynamic

coefficients, pressure distribution or other phenomena in three grid types. The

calculations with such amount of grids are made possible with the use of an

ordinary 2-core processors, 2.5 GHz computer.

4 Boundary conditions

Five types of boundaries are used in this model: Velocity inlet, pressure outlet,

symmetry, interface and wall. Dimensions of numerical domain are considered

big enough to damp turbulences before reaching the domain boundaries. The

boundary conditions for lower and upper surface of the aerofoil are considered as

solid wall.

The wind tunnel used for the experiments is a low subsonic low turbulence

(0·1% of turbulence intensity) wind tunnel at Loughborough University and the

dimensions of the test section are: 45cm × 45cm× 1·2m. The test section velocity

is 35ms–1.

The flow in this analysis is assumed to be steady, incompressible, and turbulent.

For modelling the viscous turbulent flow, k-ε turbulence model is applied.

SIMPLE algorithm is applied to solve these two equations. Standard wall

functions are also used for the areas close to wall (parietal distance y 40 ). For

the turbulence parameters at the inlet and the outlet, we use the method which

gives the intensity (0.1%) [2] and the viscosity ratio t 5 after tests. This

value is chosen for a stall detection reason.

In this part the velocity vectors and pressure coefficient contours are used to

provide a good understanding of the flow field.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

144 Advances in Fluid Mechanics IX

As evident in Figure 3, the pressure difference is higher and this will lead to a

stronger jet which will be explained later. The sudden changes in the rear portion

of the damage hole are because of the local stagnation of the flow. The gap in the

damage results was where the damage hole cut through the surface.

Figure 3(a) shows the velocity vectors around the damaged aerofoil from

lateral view for 0°deg incidence angle.

There is a weak jet exiting the damage hole which formed an attached wake

behind the damage hole.

Because of the pressure difference between two surfaces of the aerofoil, the

flow would like to penetrate from the lower surface of the aerofoil to the upper

surface and this flow through the damage is pushed through the damage hole and

added to the damage wake.

Figure 3(b) shows the velocity vectors of the numerical solution on the upper

surface of the aerofoil. At the rear edge of the damage hole, there are two vortex

centres, observed to be contra-rotating vortices approximately symmetrical about

the hole centreline.

These vortices are the results of interaction between jet flow exiting the

damage and free stream. The weak jet exited from the rear edge of the hole was

immediately bent over, forming an attached wakes to the upper surface. When

the incidence angle is increased higher than 0° degree, the jet no longer

immediately bent over upon exiting the hole and penetrated to the free stream.

Finally a separated region is formed between the jet and upper wing surface.

Figures 4(a) and 4(b) that are for incidence angle of 8°, show the separated

region and detached wake of strong jet.

It is noticeable that the stall angle of an undamaged aerofoil is normally

12°deg but for damaged aerofoil this angle shifted to 14°deg.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 145

The above numerical results for the velocity vectors on the upper surface of

the aerofoil are compared to the results of flow visualisation that was

implemented by Render et al. [4] and Mani and Render [5]. In Figure 5 that is

for damaged aerofoil, it represents experimental results. The results are presented

for incidence angles of 0°deg and 8°deg that represent weak and strong jet

respectively.

As evident, the general feature of the flow fields obtained by numerical

analysis (figures 3 and 4) are in a suitable consistency with experimental results.

It is clear in both flow visualisations that the width of the wake behind the

damage hole increases with incidence angle because of the strength of exiting jet.

There are also two large surface wakes in 8°deg incidence angle behind the

damage hole which are clear in experimental and numerical results. In both cases

they are counter rotating vortices.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

146 Advances in Fluid Mechanics IX

Some phenomena such as laminar separation and transition and thus, the

laminar bubble can not be captured by the software which are visible in the side

of the damage hole in experimental results.

The differences between experimental and numerical results are due to

numerical error associated with numerical solution.

Figure 5: Experimental flow field for damaged aerofoil at 0° deg and 8 deg

incidence angles.

In this section the quantitative results are investigated using dCL and dCD, vs α

diagrams. For a better perception and assessment of the effects of the damage;

the corresponding above results are transcribed in terms of increments of losses

of CL and CD. The increments dCL and dCD are defined from the undamaged

state as follows:

dC L C L damaged C L undamaged

dC D C D damaged C D undamaged

undamaged model was done. The results obtained were found to agree

reasonably well with those found in the literature [2].

This numerical procedure allows us to capture the stall phenomenon, for

instance, the lift coefficient reaches a maximum value of 1.20 for an angle of

stall of 12°. Also, the angle of zero lift is close to -2.5°, which is consistent with

this non symmetric profile. The coefficient of drag reaches a minimal value of

0.016 for the angle 0°.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 147

Now we consider the damaged aerofoil to explain the influence of the damage

on the aerodynamic characteristics of the aerofoil and the effect of both diameter

and chordwise position of damage expressed only in increments of lift and drag

coefficients as explained above.

In figures 6 and 7, we present, simultaneously, the influence of both diameter

and chordwise position of damage with the experimental and numerical curves.

One can see an increase of the CD over most of the incidence range except close

to the stall angle where part of the jet goes through the hole. The strength of the

jet is in direct relation with incidence angle.

0,15

0,1

0,05

0

-10 -8 -6 -4 -2

-0,05 0 2 4 6 8 10 12

dCL

-0,1

-0,15

-0,2

-0,25

-0,3

-0,35

Incidence

(experimental and numerical solutions).

0,06

0,05

0,04

dCD

0,03

0,02

0,01

0,00

0 2 4 6 8 10 12 14

Incidence

results).

There are two mechanisms for the drag increase. For small angles of

incidence, the attached jet increases friction drag while for higher angles of

incidence, the strong jet forms a separated wake which increases form drag. An

additional pressure drag is produced by the damaged hole which creates a

positive pressure increment on the wing internal surface.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

148 Advances in Fluid Mechanics IX

Results also show a reduction in the amplitude of CL over the entire range of

positive incidence angle. This reduction in CL is due to the hole through flow

which affects the distribution of pressure at the upper surface.

Lift loss for the quarter chord damage is higher than the mid-chord damage.

This is expected because when we approach the leading edge; the suction

pressure on the upper surface is strongly reduced affecting consequently the lift

coefficient. The drag increase is higher for the quarter chord damage, where the

chord wise extent of the wake is greater than seen for the mid-chord.

For the effect of diameter, results show that lift increments dCL against

incidence for different values of the diameter and over the positive incidence

range ( > -2.5°), an increase of the hole size results in a decrease of the lift

coefficient.

This is expected because a larger damage size allows a greater through flow,

and perturbs even more the pressure distribution at the upper surface. Increasing

the hole diameter changes the jet shape from weak-jet to strong-jet.

The drag also increases with hole diameter over the entire incidence range.

Indeed, an increase of the diameter increases the wake area size. In order to

validate the numerical results and to assess that the numerical procedure is

performed truly, the experimental coefficients are also required.

The aerodynamic coefficients for damaged aerofoil from numerical solution

are compared to experimental coefficients of circle damaged aerofoil that were

available in [2] and [3]. It is observed that the trends of experimental and

numerical curves in the two figures are so similar to each other and there is a

very good consistency between the curves. In the worst case, the difference

between the values of CL and CD at the high incidence close to the stall is less

than 10%. In this incidence angle, the numerical error is at the highest value

because of stall phenomenon and causes this difference.

The difference between numerical and experimental results that is observable

in figure 6 is because of numerical error. Generally, the results show that the

numerical solution has no major error in modelling and solving the equations.

The constant difference between two curves is originated from numerical

sources such as turbulence modelling, boundary-layer capturing and truncation

error. It can be concluded from this diagram that numerical solution has been

performed truly and physical phenomena are correctly captured.

8 Conclusion

In this paper the various damages have been studied comprehensively by

numerical investigation and the results have been validated using the relevant

experimental results. The multi-block grid allows us to reduce time consumption

and enhances consistency and accuracy. It is concluded that the damages (circle

damage) give a reduction in lift and an increase in drag. These variations depend

on the incidence angle and the diameter (jet exiting from the damage hole). The

numerical simulation gives accurate results comparatively to experimental

results.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 149

References

[1] Ball, R. E., The Fundamentals of Aircraft Survivability Analysis and Design,

2nd ed., AIAA Education Series: Reston, VA, 2003.

[2] Irwin, A. J. & Render, P. M., The influence of mid-chord battle damage on

the aerodynamic characteristics of two-dimensional wings, The Aeronautical

Journal, 104(1033), pp.153–161, 2000.

[3] Irwin, A. J., Render, P. M., McGuirk, J. J., Probert, B. & Alonze, P. M.,

Initial investigation into the aerodynamic properties of a battle damaged

wing. 13th AIAA Applied Aerodynamics Conference, AIAA Paper No. 95-

1845, 1995.

[4] Render, P. M., De Silva, S., Walton, A. J. & Mani, M., Experimental

investigation into the aerodynamics of battle damaged airfoils, Journal of

Aircraft, 44(2), pp.539–549, 2007.

[5] Mani, M., & Render, P. M., Experimental investigation into the

aerodynamic effects of airfoils with triangular and star shaped through

damage,” AIAA Paper 2005-4978, June 2005.

[6] Djellal, S., & Ouibrahim, A., Aerodynamic performances of battle-damaged

and repaired wings of an aircraft model”, Journal of Aircraft, 45(6),

pp.2009–2023, 2008.

[7] Saaedi, M., Ajalli, F. & Mani, M., A., Comprehensive numerical study of

battle damage and repairs upon the aerodynamic characteristics of an

aerofoil, The Aeronautical Journal, 114(1158), August 2010.

[8] Djellal, S., Effects of Battle Damage Repairs on Aerodynamic of a wing,

Sabbatical leave report, Aeronautical and Automotive Engineering

Department, Loughborough University, 2004.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

This page intentionally left blank

www.engbookspdf.com

Advances in Fluid Mechanics IX 151

an elastic cylinder in an axial flow

Z. G. Liu1, Y. Liu1 & J. Lu2

1

Department of Mechanical Engineering,

Hong Kong Polytechnic University, China

2

College of Engineering, City University of Hong Kong, China

Abstract

The fluid-structure interaction for an elastic cylinder in an axial flow was studied

numerically with the ALE method and the LES turbulence model. The

commercial CFD software Fluent is used as the fluid solver and the Euler-

Bernoulli beam solver is embedded into Fluent by its user defined functions

(UDF). Two types of cylinder are included in this study. The motion for the first

type of cylinder is constrained in one lateral direction and for the second in two

lateral directions. The two types of cylinder are both released from their

equilibrium state. When the stiffness is kept large enough, only weak oscillatory

is induced by the flow. However, the motion of the cylinder induced by the flow

may become unstable in the form of either buckling or oscillatory, when the

stiffness becomes small enough. In this study, it is found that with the same

simulation parameters the first type of cylinder is buckled and the second has an

oscillatory. When buckled, the cylinder is located at a new state with weak

oscillatory. The oscillatory after the instability has much larger amplitude than

that before the instability.

Keywords: CFD, fluid-structure interaction, fluidelastic instability, buckling,

flutter.

1 Introduction

One of the most classical subjects in fluid mechanics is the flow over cylinder,

by which many interesting and significant phenomena are discovered such as

Karman vortex streets [1] and also by which, for example, one can get the Kutta-

Joukowski theorem, one of the most basic theorems in aerodynamics [2].

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120141

www.engbookspdf.com

152 Advances in Fluid Mechanics IX

However, this is almost for a rigid cylinder and the flow past a deformable

cylinder attracts more and more interest of researchers due to not only its

academic importance but also a lot of applications in engineering such as nuclear

station design, building and so on. In this case, the fluid-structure interaction

(FSI) will arise. FSI is a big topic and includes many aspects, nevertheless in this

paper we just focus on a slender deformable cylinder subjected to axial flow.

This problem has been studied by many researchers and a good recent review

was conducted by Wang and Ni in [3], where they also reviewed other FSI

systems besides the system in this paper. In 1966, Paidoussis [4] developed a

theory to deal with the dynamics for a single cylinder subjected to axial flow

based on the Lighthill theory [5], which was used to get the force related to

added mass, and Taylor’s theory [6], which was applied to calculate viscous

force imposed on the cylinder. The two kinds of force related to fluid flow and

other kinds of force all were imposed on the cylinder regarded as an Euler-

Bernoulli beam in his theory [4]. Based on his theory, Paidoussis [4] concluded

that the stability of the system is determined partly by the dimensionless

velocity, the definition of which will be written in the following, i.e. the flow

damps the vibration for small dimensionless velocity while the cylinder may

become instable, either in the form of buckling or flutter, when this parameter

becomes large enough. He also conducted experiment to check his theory [7].

Taking gravity into account and modifying frictional force, Paidoussis also

developed his theory further [8], however still a linear theory, which was mended

further by Lopes et al. [9] into a nonlinear one for cantilevered cylinder in axial

flow. The basic conclusion based on this nonlinear theory and that based on the

linear theory have no crucial differences [10, 11]. In addition, the idea of

Paidoussis’s deriving his theory can be extended to several cylinders subjected to

axial flow [12–14].

When the flow velocity is small, the cylinder could vibrate but with small

amplitude. This kind of vibration, often called sub-critical vibration, is related to

the perturbation of flow field [15, 16]. Some former researchers used semi-

empirical methods to predict the amplitude of this kind of vibration [17–19]. But

one must have the loading distribution available from experiments in order to

adopt those methods in practice.

It is very manifest that one should consider both the fluid flow field and the

structure dynamics when he deals with FSI. In some simple cases, one can get an

analytic solution for the FSI system [20]. But unfortunately, on the other hand, in

the most cases one can not get a fluid flow field solution to give enough

information to the structure, even if the structure is not of complexity, e.g. the

FSI system in this paper, a cylinder subjected to axial flow. Even in the theories

mentioned above, the flow field is actually obtained based on some assumptions.

Thus in this paper, we simulate numerically the FSI system shown in fig. 1,

namely solve the fluid flow field in the arbitrary Lagrange-Euler (ALE) frame

with CFD and the cylinder dynamics equation numerically. The paper is divided

into four parts. The model and governing equations will be introduced in the

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 153

second part, followed is the third part, where the numerical details will be

illuminated. In the last part, we will discuss the results and make a brief

conclusion.

The FSI system is shown in fig. 1, where D denotes the outer diameter of the

cylinder. The cylinder clamped at its two ends, 20D long with its inner diameter

0.88D, is located in the fluid flow confined by a cylindrical pipe that is the same

long as the cylinder with 4D diameter. The fluid could flow from one end to the

other, inducing the vibration of the cylinder. In this paper, we study two cases for

the model, in the first of which the cylinder is constrained only to vibrate in one

lateral direction, while in the second of which the cylinder can vibrate in two

lateral directions.

For the different cases respectively, the same parameters are adopted in the

simulation. The main parameters are listed in table 1, where the dimensionless

velocity υ and the mass ratio β are defined by eqns. (7) in section 4. The

Reynolds number Re is based on the hydraulic diameter, which is 3D, and the

inlet flow velocity υ0.

Case 1 2

4

Re 8×10

υ 6.0173 8.8997 6.0173 8.8997

β 0.4699 0.4699 0.4699 0.4699

First, it should be noted that the fluid flow velocity here is small and there is no

need to consider the compression of the fluid. Therefore the incompressible

Navier-Stokes (N-S) equations are one of the bases. Generally speaking, the fluid

domain deforms in the process of simulation due to the deformation of the

interfaces between the fluid and structure domains. To take into account the

deformation of the fluid domain essentially, one method for CFD is rewriting the

N-S equations as arbitrary Lagrange-Euler form, i.e. ALE N-S equations [20, 21]:

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

154 Advances in Fluid Mechanics IX

υ 0

υ (1)

t ( υ υˆ ) υ p υ

2

where t, υ, p, ρ and μ are time, fluid velocity, pressure, density and viscosity

respectively; υ refers to mesh node velocity, which related to the deformation of

the fluid domain. One can find the detailed derivation in [20]. In eqns. (1), the

flow variables are solved in Euler frame as usual while the mesh nodes are

regarded in Lagrange frame. This is why eqns. (1) are called ALE N-S equations.

It is easily seen that eqns. (1) become the common N-S equations if the mesh

node velocity υ 0 . The large eddy simulation (LES) model [22, 23] is adopted

to deal with the turbulence and the Smagorinsky-Lilly model [22, 23] is utilized

as the subgrid-scale model.

The cylinder is considered as an Euler-Bernoulli beam and currently the axial

loading and gravity are ignored here. The coordinates system is shown in fig. 1

and right-hand. The dynamic equation is [24]

2u 4u

b A EI q( z, t )

t 2 z 4 (2)

with u(z) being the displacement of cross sections in beam, EI, A, ρb being the

modulus of flexural rigidity, cross-sectional area, density of the beam,

respectively, q(z, t) being the loading on beam and calculated after the fluid field

is solved. Eqn. (2) is just for the vibration of cylinder in one plane (for example

x-z plane) and the dynamic equation for the other plane is ignored here, because

they have the same form. Eqn. (2) can be solved numerically by finite element

method (FEM), and the numerical details are explained in the following part.

The boundary conditions at two ends of the cylinder are both clamped as

mentioned above, which give

u u

u z L / 2 u zL / 2 0

z z L / 2 z zL / 2 (3)

with L = 20D being the length of the cylinder (see fig. 1).

In addition, the mesh is remeshed at each simulating step by spring analogy

[23], in which the nodes of mesh are considered as being connected by fictitious

springs whose stiffnesses are concerned with the locations of nodes. In this

paper, the fictitious spring constant between two nodes is assumed being inverse

proportion of their distance [23].

3 Numerical approach

The simulation on FSI actually needs to treat three distinct fields coupled

together [25]. The first field is fluid flow field governed by N-S equations, the

second is related to the structure deformation and the last one refers to the mesh

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 155

motion. These three coupled fields give rise to three equations sets still coupled

together after discretization, making a great challenge for numerical simulation.

In the past years, to overcome this challenge, many researchers [25–28] proposed

many available numerical approaches, which, however generally speaking, can

be classified roughly into two major categories, monolithic [26] and partitioned

[25, 27, 28]. The monolithic approaches do not uncouple three numerical

equations sets and solve them simultaneously as one, while the partitioned

approaches solve them in sequence. The monolithic approaches have better

numerical stability [27] but cost more computational resources than the

partitioned ones. In practice, one of major difficulties in adopting the monolithic

approaches is that the programming codes for them should usually be rewritten

and can not reuse efficiently the already existing CFD or structure solvers.

Therefore, the partitioned approaches are still popular and applied in this study.

The applied partitioned scheme is shown in fig. 2 and the same as CSS in

[28]. The commercial software Fluent is utilized as the CFD solver, in which the

ALE N-S equations are discretized by finite volume method (FVM) [23], and the

structure codes by finite element method [29] based on eqns. (2) and (3) are

embedded into the former by its providing user-defined functions (UDF) [23],

some of which provide the access to controlling the motion of fluid domain mesh

in simulation. The algorithm of updating mesh has been explained above and is

implemented in Fluent. Firstly, the flow field is calculated by Fluent and

sequentially obtained is the loading exerted on the cylinder (regarded as a beam),

by which the structure codes determine the deformation of the cylinder, forming

new boundary for the fluid domain. Then Fluent updates the mesh and a new

cycle begins if needed. It should be noted that the fluid flow in all cases is

simulated with turbulence model.

For the flow simulation, the first order implicit method for temporal

discretization and the bounded central differencing scheme for spatial

discretization are adopted with the PISO method for pressure-velocity

coupling [23]. For turbulent flow, the LES model is utilized with Smagorinsky-

Lilly as subgrid-scale model [22, 23].

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

156 Advances in Fluid Mechanics IX

Galerkin procedure of FEM [29], it with eqn. (3) can be discretized as one

ordinary differential equations set

Ku Q

Mu (4)

where M and K are global mass and stiffness matrices respectively;

u [u1 , u 2 ,, u N ]T , a function of time t and z, is generalized displacements

vector of N nodes including two end nodes. For one certain node, its

generalized displacement consists of its displacement and rotation, i.e.

u i [vi , i ] (5)

and K are assembled by local mass and stiffness matrices respectively. One

can find the details on these local mass and stiffness matrices and how to

assemble them to global ones as well as how to get Q in [29 , 30] . Eqn . (4) is

solved by the generalized-α method, an extension version of HHT-α and

WBZ-α method, having the second-order accuracy and being unconditional

stable if its parameters are set appropriately [31].

The loading here includes pressure and frictional force, the same as proposed

by Paidoussis [4, 16] except the ignored axial force, i.e.

q i ( p ij ij ) n j dl . (6)

C

where qi is the loading in eqn. (2) with i = x, y, τij is the deviatoric stress tensor

and nj refers to the normal vector of the cylinder, and C is its lateral perimeter.

For convenience, following Paidoussis [4, 16], we define dimensionless velocity

υ, the mass ratio β, the dimensionless time τ and displacement by

At At t EI

0 L , , 2 ,

EI At b Ae L b Ae At

u x ( z) u y ( z)

dx , dy (7)

D D

where υ0 and ρ are the flow inlet velocity and fluid density respectively; L, ρb, At,

and Ae are respectively the length, density, total cross-sectional area and effective

cross-sectional area of the cylinder; dx and dy are the dimensionless displacement

in x direction and y direction respectively with ux and uy being the corresponding

displacement. Note that Ae is actually A in eqn. (2) and different from At since the

cylinder is hollow.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 157

the case where only one direction vibration is calculated and case 2 to the other

one where the vibrations in two directions are both calculated. For different

cases, the cylinder is located initially at its equilibrium state and only induced by

fluid flow to vibrate. By increasing or decreasing stiffness of the cylinder, we

can make the dimensionless velocity smaller or lager and thus focus on the effect

of the latter for generalizing, by which the effect of the former can be easily and

clearly concluded. In this paper, the effect of β is not discussed.

Fig. 3 shows the displacement histories of two cross sections of the cylinder

in x direction for the case 1, where ζ is dimensionless length based on the

diameter of the cylinder and ζ = 0, 20 means two ends of the cylinder

respectively. These two cross sections have the almost most maximum

displacement and thus the following results also focus on them.

It can be found from fig. 3 (a) that the oscillation amplitudes are small, about

maximum 0.01D with D diameter of the cylinder, when the dimensionless

velocity υ equals 6.0173. This kind of oscillation is almost around the initial

equilibrium position of the cylinder and attributed to the turbulence as we

mentioned above and called sub-critical vibration. However, when υ is increased

to 8.8997, or equivalently the stiffness of the cylinder is decreased, as shown in

fig. 3 (b) one dramatic vibration is induced distinct from that shown in fig. 3 (a).

At the beginning, the oscillations with small amplitudes are induced till τ = 1.0 as

they are when υ = 6.0173, after which the displacements rapidly increase to a

large value about maximum 0.6D within a short time and then will never do

visibly. Consequently, the cylinder oscillates around a new position with an

approximate maximum amplitude 0.025D lager than that when υ = 6.0173. It

indicates that the cylinder is buckled at a new position. This phenomenon is one

type of fluidelastic instability and in particular discussed for the cylinder

subjected to axial flow by Paidoussis [4, 7, 16].

We now move on to the case 2, where the cylinder can vibrate in two

directions. Fig. 4 shows the vibrations of two same selected cross sections in two

directions when υ = 6.0173. The oscillating amplitudes in two directions are also

small, about maximum 0.05D in x direction and 0.03D in y direction. No matter

in x and y directions, the oscillating is almost around the initial equilibrium

position of the cylinder. This is similar to that shown in fig. 3 (a) and also this

oscillation should be related to the turbulence. Nevertheless, when υ = 8.8997,

the difference between case 1 and case 2 is apparent. Fig. 5 shows the vibrations

in two directions when υ = 8.8997. Initially, the cylinder vibrates around its

equilibrium position and then deviates rapidly from this position. The deviation

is followed by an oscillation with the amplitude approximately maximum 0.1D

in x direction and 0.2D in y direction both much larger than that after rapid

increasing of the displacement shown in fig. 3 (b). This oscillation is different

from the initial oscillation that should be only about 10-3 to 10-2D [16] and thus

could not be mainly associated with the turbulence but due to another kind of

fluidelastic instability called flutter. Comparing fig. 3 (b) and fig. 5, one can

further find that the time of the displacement rapidly increasing onset in the

former is after that in the latter, together with the fact that the buckling usually

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

158 Advances in Fluid Mechanics IX

occurs at a smaller value of the dimensionless velocity than the flutter [4, 16]

concluding that the FSI system here is more stable when the cylinder is

constrained only to vibrate in one direction than when it can vibrate in two

directions.

(a)

(b)

Figure 3: Dimensionless x-displacements of two cross sections of the

cylinder in the case 1 when (a) υ = 6.0173 and (b) υ = 8.8997.

The buckling is not predicted by Paidoussis theory [4] for his considering a

specific model similar to the model in this paper, but by recent nonlinear theory

[32], the buckling is possible for a similar clamped-clamped system. The value

8.8997 of the dimensionless velocity for buckling (in case 1) or flutter (in case 2)

is different from that of fluidelastic instability onset suggested by Modarres-

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 159

Sadeghi et al. [32], which is about 2π for buckling and 21.6 for flutter. It should

be noted that the numerical method in this paper is much difficultly applied to

predict the value of the dimensionless velocity after which the fluidelastic

instability occurs, since it is impossible to simulate sufficient cases for different

dimensionless velocities due to long time cost in numerical simulation. The

value 8.8997 may not be one of fluidelastic instability onsets for the FSI system

in this paper. Another reason for this difference should be attributed to the

difference between the FSI systems in this paper and [32]. However, the results

here are qualitatively consistent with those in [4, 16, 32].

(a)

(b)

cross sections of the cylinder in the case 2 when υ = 6.0173.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

160 Advances in Fluid Mechanics IX

(a)

(b)

cross sections of the cylinder in the case 2 when υ = 8.8997.

In summary, for the FSI system in this paper, the numerical results shows that

only small oscillation of the cylinder is induced when the dimensionless velocity

is small enough while the fluidelastic instability occurs when it is large enough.

However, due to the fluidelastic instability, the cylinder is buckled when it can

vibrate only in one direction and oscillates with large amplitude when it can

vibrate in two directions. This is qualitatively consistent with the conclusions of

[4, 16, 32]. By the definition of the dimensionless velocity (see eqn. (7), it is can

be concluded that when the stiffness is kept large enough, only weak oscillatory

is induced by the flow, however, the FSI system becomes unstable if it is small.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 161

Acknowledgement

Support given by the HKPolyU/AREVA collaboration under grant No.RPQ4 is

gratefully acknowledged.

References

[1] White F.M., Viscous Fluid Flow, McGraw-Hill: New York, pp. 4-15, 2006.

[2] Anderson J.D., Fundamentals of Aerodynamics. McGraw-Hill: New York,

pp. 232-239, 2001.

[3] Wang L. and Ni Q., Vibration of Slender Structures Subjected to Axial

Flow or Axially Towed in Quiescent Fluid. Advances in Acoustics and

Vibration, 2009, pp. 1-19, 2009.

[4] Paidoussis M.P., Dynamics of Flexible Slender Cylinders in Axial Flow,

part 1: theory. Journal of Fluid Mechanics, 26, pp. 717-736, 1966.

[5] Lighthill M.J., Note on the Swimming of Slender Fish. Journal of Fluid

Mechanics, 9, pp. 305-317, 1960.

[6] Taylor G., Analysis of the Swimming of Long and Narrow Animals.

Proceedings of the Royal Society of London, Series A, Mathematical and

Physical Science, 214, pp. 158-183, 1952.

[7] Paidoussis M.P., Dynamics of Flexible Slender Cylinders in Axial Flow,

part 2: experiments. Journal of Fluid Mechanics, 26, pp. 737-751, 1966.

[8] Paidoussis M.P., Dynamics of Cylindrical Structures Subjected to Axial

Flow. Journal of Sound and Vibration, 29, pp. 365-385, 1973.

[9] Lopes J.L., Paidoussis M.P. and Semler C., Linear and Nonlinear Dynamic

of Cantilevered Cylinders in Axial Flow, part 2: The Equation of Motion.

Journal of Fluids Structures, 16, pp. 715-737, 2002.

[10] Paidoussis M.P., Grinevich E., Anamovic D. and Semler C., Linear and

Nonlinear Dynamic of Cantilevered Cylinders in Axial Flow, part 1:

Physical Dynamics. Journal of Fluids Structures, 16, pp. 691-713, 2002.

[11] Semler C., Lopes J.L., Augu N. and Paidoussis M.P., Nonlinear Dynamic

of Cantilevered Cylinders in axial flow, part 3: Nonlinear Dynamics.

Journal of Fluids Structures, 16, pp. 739-759, 2002.

[12] Chen S.S., Vibration of Nuclear Fuel Bundles. Nuclear Engineering and

Design, 35, pp. 399-422, 1975.

[13] Paidoussis M.P. and Suss S., Stability of a Cluster of Flexible Cylinders in

Bounded Axial Flow. Journal of Applied Mechanics, 44, pp. 401-408,

1977.

[14] Paidoussis M.P., The Dynamics of Clusters of Flexible Cylinders in Axial

Flow: Theory and Experiments. Journal of Sound and Vibrations, 65, pp.

391-417, 1979.

[15] Paidoussis M.P. and Curling L.R., An Analytical Model for Vibration of

Clusters of Flexible Cylinders in Turbulent Axial Flow. Journal of Sound

and Vibrations, 98, pp. 493-517, 1985.

[16] Paidoussis M.P., Fluid-Structure Interactions: Slender Structure and Axial

Flow, Vol. 2, Elsevier: London and California, pp. 787-877, 2004.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

162 Advances in Fluid Mechanics IX

Displacement in Parallel Turbulent Flow. Nuclear Science and

Engineering, 38, pp.63-69, 1969.

[18] Reavis J.R., WVI-Westinghouse Vibration Correlation for Maximum Fuel-

Element Displacement in Parallel Turbulent Flow. Transactions of the

American Nuclear Society, 10, pp. 369-371, 1967.

[19] Gorman D.J., An Analytical and Experimental Investigation of the

Vibration of Cylindrical Reactor Fuel Elements in Two-Phase Parallel

Flow. Nuclear Science and Engineering, 44, pp. 277-290, 1971.

[20] Wang X.D., Fluid-Solid Interactions: Analytical and Computational

Approaches, Elsevier: Amsterdam and Oxford, pp. 163-169, 2008.

[21] Stein E., De Borst R. and Hughes J.R., (eds). Encyclopedia of Com-

putational Mechanics, Vol. 1 (Chapter 14). Arbitrary Lagrangian-Eulerian

Methods, ed. Donea J. etc, John Wiley and Sons, Ltd., pp. 1-25, 2004.

[22] John V., Large Eddy Simulation of Turbulent Incompressible Flows.

Springer-Verlag: Berlin and Heidelberg, pp. 21-62, 2004.

[23] ANSYS FLUENT 12.0 Documentation, ANSYS, Inc.

[24] Chen Y., Vibrations: Theoretical Methods. Addison-Wesley Publishing

Company, Inc.: Reading and Massachusetts, pp. 204-208, 1966.

[25] Piperno S., Farhat C. and Larrouturou B., Partitioned procedures for the

transient solution of coupled aeroelastic problems, part 1: Model problem,

theory and two-dimensional application. Computer Methods in Applied

Mechanics and Engineering, 124, pp. 79-112, 1995.

[26] Blom F.J., A monolithical fluid-structure interaction algorithm applied to

the piston problem. Computer Methods in Applied Mechanics and

Engineering, 167, pp. 369-391, 1998.

[27] Matthies H.G. and Steindorf J., Partitioned but strongly coupled iteration

schemes for nonlinear fluid-structure interaction. Computers and

Structures, 80, pp. 1991-1999, 2002.

[28] Farhat C. and Lesoinne M., Two Efficient Staggered Algorithms for the

Serial and Parallel Solution of Three-Dimensional Nonlinear Transient

Aeroelastic Problems. Computer Methods in Applied Mechanics and

Engineering, 182, pp. 499-515, 2000.

[29] Zienkiewicz O.C., Taylor R.L. and Zhu J.Z., The Finite Element Method:

Its Basis and Fundamentals. Elsevier Butterworth-Heinemann: Oxford and

New York, pp. 54-95, 2005.

[30] Zienkiewicz O.C. and Taylor R.L., The Finite Element Method for Solid

and Structural Mechanics. Elsevier Butterworth-Heinemann: Oxford and

New York, pp. 278-320, 2005.

[31] Chung J. and Hulbert G.M., A Time Integration Algorithm for Structural

Dynamics with Improved Numerical Dissipation: The Generalized-α

Method. Journal of Applied Mechanics, 60, pp. 371-375, 1993.

[32] Modarres-Sadeghi Y., Paidoussis M.P. and Semler C., The Nonlinear

Behavior of a Slender Flexible Cylinder Pinned or Clamped at Both Ends

and Subjected to Axial Flow. Computers and Structures, 85, pp. 1121-

1133, 2007.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 163

vibration of a pivoted cylinder using the

spectral element method

H. Gun Sung1, H. Baek2, S. Hong1 & J.-S. Choi1

1

Maritime and Ocean Engineering Research Institute (MOERI; formerly

KRISO), KORDI, Daejeon, Korea

2

Department of Mathematics, MIT, Cambridge, USA

Abstract

The present paper describes preliminary results of numerical simulation of VIV

(vortex-induced-vibration) phenomenon around a pivoted circular cylinder

subject to steady flow. The present flow model is based upon the Navier-Stokes

equations with velocity-pressure formulation. The governing equations are

solved through the Spectral Element Method (SEM), which possesses the

property of high-order spatial accuracy as proposed by Karniadakis and Sherwin

in their book Spectral/hp Element Methods for CFD. The solution procedure and

characteristic aspects of the present modelling and numerical method are briefly

stated. The coupling method of the body motion with the flow problem is

restated from the viewpoint of the present problem. A series of numerical

estimation of VIV flow characteristics have been carried out for varying

parameters of the problem such as the reduced velocity and damping parameter,

etc.

Keywords: vortex-induced vibration, pivoted cylinder, spectral element method,

coupling of body motion with flow.

1 Introduction

In general, VIV (vortex-induced-vibration) problem is focused upon reduction

and suppression of VIV as noted in Blevins [2] and Sarpkaya [3]. As indicated

by Raghavan [4], and Bernitsas and Raghavan [5], however, it seems to be

probable to extract considerable energy from VIV phenomena. For example,

Professor Bernitsas has developed an energy extraction device known as

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120151

www.engbookspdf.com

164 Advances in Fluid Mechanics IX

ocean and river current energy to electricity (Bernitsas et al. [6]).

MOERI/KORDI is also developing its own energy extraction device which

utilizes a pivoted circular cylinder subject to background current flows.

The present paper consists as follows. The physical problem under

consideration is introduced. Flow model and governing equation are briefly

explained. Numerical method and solution procedure are addressed. Some of

numerical results of a specified configuration are presented.

As shown in Fig. 1, a submerged cylindrical body is pivoted to a column and

exposed to VIV due to background fluid flow such as ocean currents. As stated

above, the present device is conceptually associated with the earlier development

done by Prof. Bernitsas’ group in the University of Michigan.

cylinder subject to steady flow.

Coordinate system and modeling of present problem is shown in Fig. 2. Rotation

vector of the body can be denoted by (t ) = (0,0, (t )) for the present case of

2D. So, the angular velocity and acceleration can be obtained by time derivative

of the rotation vector as (t ) and (t ) , respectively. We also assume the flow

domain has rectangular shape, and there is no free surface.

We describe the present flow problem by using the viscous incompressible

flow model. For the present problem in which we need to describe the grid

movement due to vortex induced body motion, we can apply ALE (Arbitrary

Lagrangian-Eulerian) framework. Hence, the corresponding set of governing

equations can be modified as follows.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 165

u

N u , w p Lu f in V (t )

1

(1)

t

u 0 in V (t ) (2)

where u ( x , t ) (u, v, w) is the velocity vector, w( x , t ) the ALE mesh velocity

vector, p ( x , t ) the pressure field, the density of the fluid, the kinematic

viscosity of the fluid, and f the body force that can include the effect of non-

inertial reference frame. The underlying Cartesian coordinates is denoted by the

spatial coordinates x and by the time coordinate t . We can define as

N (u , w) (u w) u and L(u ) u (u )T . So, the partial time

derivative of the velocity is taken for a given fluid particle or a pseudo-particle in

general. It is also assumed that the fluid domain, V (t ) can be time-varying for

generality of the formulation. With ALE concept, the partial time derivative in

the above is taken on the mesh frame under consideration.

follows.

u ub in b (t ) (3)

where ub (t ) (r rCR ) with rCR ( xCR , yCR , zCR ) to denote the position

vector of the center of rotation, i.e. the pinned position of the rod.

In the present paper, we apply the boundary conditions at truncation

boundaries as follows.

u U , w 0 in x x L (4)

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

166 Advances in Fluid Mechanics IX

w 0, s u (u )T n 0 in z z L , z H (5)

u U , s u (u )T n 0 in x x L (6)

where U denotes the outflow velocity, and x xL , x xR , z z L , z z H mean

the left, right side and low, high side of the truncated rectangle of the flow

domain. The vectors s and m (pointing to z-axis in 2D problems) are an ortho-

normal vector pair lying on the tangent plane, and related to the unit normal

vector as n s m . Hence, the vector pair, ( s , m, n ) constitutes the ortho-

normal curvilinear coordinates.

With the ALE methodology, the motion of the mesh nodes, X is described

as

dX

w (7)

dt

with an operator equation for the mesh velocity vector,

Εw 0 (8)

where Ε stands for the elliptic operator which governs the grid motion. Its type

is chosen depending upon the problem in consideration. For example, the

Laplace equation was utilized by Ho and Patera [7], and Robertson et al. [8]. On

the other hand, Rabier and Medale [9] adopted a Lagrangian-Eulerian kinematic

method by the way of an elastic problem for the mesh displacement field rather

than a standard ALE formulation. A similar approach can be found in Bouffanais

and Deville [10].

The boundary condition on the ALE mesh motion is also to be carefully

specified. On the free-slip boundary, ALE mesh nodes move with the fluid

particle as follows.

w u (u n )n (u s ) s (u m)m in s (9)

This guarantees the velocity compatibility on the free-slip boundary. The

Dirichlet-type boundary specifies the velocity vector and the corresponding

boundary is stationary in space and time. So, we may apply the following

constraint.

w 0 in D (10)

In the case of 2D problem, the equation of body motion can be written as

follows.

J b Cb K b Text (t ) (11a)

where denotes the angle of rotation of the body. The coefficients J b , Cb , K b

are the moment of inertia, damping factor, and stiffness of rotation, respectively.

The external moment, Text can be obtained from the equation.

Text (t ) e3

r rCG pn u u T n ds

Sb (t )

(12)

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 167

where rCR means the position vector of the center of rotation, i.e. the pinned

position of the rod, and e3 (0,0,1) for the present case. For simplicity of the

derivation and notation, the equation (11a) can be normalized as follows:

20 b 02 Text* (t ) (11b)

with definitions such as 0 K b / J b , b Cb / 20 J b , Text

*

Text (t ) / J b .

Based upon the proposed values of preliminary design, some variations of the

parameters such as the length of the connection rod, mass moment of inertia,

damping factor, and stiffness are compared with in this study by numerical

simulation.

In order to couple the body motion with the fluid flow, we need to transfer the

velocity and acceleration on the body surface to the flow solver. Thus, we can

use the following relationship.

ub (t )( y yCR , x xCR ) in b (t ) (13)

where we use (t ) (0,0,) . The acceleration is written as follows.

ab (t ) (r rCR ) (t ) (t ) (r rCR ) (14)

Thus, in the present case we have

ab (t )( y yCR , x xCR ) (t ) 2 ( x xCR , y yCR , ) in b (t ) (15)

For the present case of rotating circular cylinder, the body boundary can be

defined as

x xCR Rr cos 2 y yCR Rr sin 2 Rc2 (16)

where the parameter is the angle of the point on the cylinder surface, R r

means the distance from the center of rotation and center of the cylinder, and Rc

the cylinder radius.

For numerical solution of the ALE N-S equations, we utilized the high-order

splitting scheme for pressure and velocity coupling by using stiffly-stable time

integrators. It was originally developed by Professor G.E. Karniadakis

(Karniadakis and Sherwin [1]), which is utilized in this study with some

modification. Detailed procedure is thus omitted and a brief introduction is

given.

The high-order splitting scheme is applied as follows.

~ J 1 J 1

u q u n q t q N u n q , wn q f n 1 in n

i i

(17)

q 0 q 0

~

~ ~ t

u u p n 1 in n (18)

~

~

0u n1 u tL u n1 in n (19)

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

168 Advances in Fluid Mechanics IX

The superscript refers to the time level index, and J i the order of the time

integration, and 0 , q and q the coefficients for stiffly stable time integration.

Readers are referred to Karniadakis and Sherwin [1] for values of the integration

coefficients for different orders. The time step size, t is fixed throughout the

time marching in this study.

The primary reason for using the present time integration scheme is due to

necessity of attaining the high-order accuracy and robustness in time. The stiffly-

stable time integration scheme is known to be very accurate and stable.

Particularly in SEM, the condition number is very large compared with other

discretization schemes and so the conventional time integration scheme such as

the Adams-Bashforth and Runge-Kutta methods may give rise to numerical

instability.

The present study is based upon the previous development known as Nektar

which was originally due to Prof. Karniadakis at Brown University, USA. Thus

in this paper, we can skip most of the details and just cite relevant references

such as Karniadakis and Sherwin [1]. Etc. It should be noted that the grid motion

is directly obtained by using ALE concept which is implemented on the

framework of NEKTAR.

Coupling of the flow dynamics with body motion is realized mainly through time

marching procedure. In the above, we mentioned the type of time-stepping for

fluid flow. So, the remaining part is the equation of body’s rotational motion,

and data transfer and connection between fluid part and structure part, which are

discussed below.

The conventional method is based upon the Newmark-family time marching

as in Papaioannou et al. [11] and Placzek et al. [12], etc. When we are interested

in explicit time marching, any type of time integration scheme can be utilized in

principle only if it guarantees numerical stability. However, the practical issue

seems to be efficiency and accuracy as well; i.e. a choice is preferred when the

type of time marching algorithm for structure part does not deteriorate the level

of accuracy of solution method of fluid part, particularly in time-axis. The

present method for this FSI problem is briefly explained in the below.

At first, the angular motion of the pivoted cylinder can be described by

n 1 n 1 n 1

J C K

n

b b Tb ext (20)

where the superscript denotes the time step, while the subscript means sub-

n

iteration step. Hence, Text stands for the hydrodynamic torque that can be

calculated by using the previous time step.

Within this framework, the way to describe n and n can construct the time

integration scheme and also can determine its temporal accuracy and stability.

The present choice for time discretization is a Newmark scheme which can be

written as

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 169

n 1 n t n

2

t 2

(1 2 )n 2 n 1 (21a)

n 1

n t (1 )n n 1 (21b)

It is known that the accuracy of the scheme is dependent upon the parameter

set of ( , ) . In this study, ( , ) (0.25,0.5) is used.

The whole process for each time step can be summarized as follows.

i) Solve the equation for the solid part with the last solutions of the fluid

part.

ii) Obtain the angular displacement, velocity, and acceleration using Eqs.

(21a) and (21b).

iii) Calculate the translational velocity and acceleration for the next time-

step on the body boundary and transfer those values to fluid flow solver.

iv) Solve the governing equation of fluid motion, and the corresponding

mesh field.

v) Compute the traction on the interface between and the solid and the

fluid and find the torque acting upon the body.

vi) Repeat the processes i) to v) up to the termination time.

In this section, we discuss the numerical results of the present VIV problem of a

pivoted circular cylinder for varying parameters with the developed numerical

method. A typical grid system for present VIV simulation is shown in Fig. 3. As

shown, rectangular grids are utilized, and the order of the interpolation of

spectral elements was 5 or 6 for most of the cases of simulation.

0

y

-2

-4

-5 0 5 10 15 20

x

simulation for which we have experimental data. In this case, the flow is incident

from the upstream side and the pivoting point is located downstream. The

normalized mass moment of inertia was 9.089 for the present case.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

170 Advances in Fluid Mechanics IX

displacement, velocity, acceleration, and hydrodynamic torque of the cylinder is

shown in Fig. 4. The Reynolds number was 200, the reduced velocity was 5, and

the damping factor was 0.07 for this case. Figure 5 depicts how the cylinder is

moving in the plane for the same case. Figure 6 and 7 correspond to pressure and

vorticity contour plots for this case at a specified time instant. It is noted that the

vorticity contour plot shows some discontinuities around mesh boundary because

of relatively coarse grid density in the downstream side.

0.8

0.4

-0.4

angular displacement

-0.8 Angular velocity

Angular acceleration

Hydrdynamic torque

-1.2

0 10 20 30 40 Nondim. time 50

acceleration, and hydrodynamic torque of the cylinder

( Rr / Rc 3.0 , Re 200 , U r U / 0 Rc =5, b 0.07 ).

U r U / 0 Rc =5, b 0.07 ).

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 171

U r U / 0 Rc =5, b 0.07 ).

U r U / 0 Rc =5, b 0.07 ).

performed by Choi et al. [13]. It is noted that in the experiment the cylinder is

connected to the multi-hinged bars which is also guided to a generator unit. The

detailed features of the experiment are referred to the paper cited in the above.

The figure indicates that the present numerical method under-predicts the

performance of the rotational VIV motion of the cylinder. We note that the

present value of damping factor in simulation was found by some series of decay

test in the experiment. We suppose that the present numerical method contains

appreciable level of numerical damping because the numerical results is similar

to the no load case rather than to the zero resistance case.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

172 Advances in Fluid Mechanics IX

0.50

Experiment (Resistance = 0 Ohm)

Computation (Navier-Stokes; Re=200)

Computation (Navier-Stokes; Re=2,000)

0.40

0.30

0.20

0.10

0.00

2 4 6 8 U/fD 10 12

Figure 8: VIV response for various reduced velocity ( Rr / Rc 3.0 , for the

computation, the damping factor was chosen as b 0.07 ).

Re 1.625 10 5 , while for present simulations we tested the case of 200

Re 2,000 . Since the higher Reynolds numbers cause the present numerical

method to show an unstable motion in time. As an intermediate conclusion, it is

required to utilize the strong coupling scheme as noted in Baek [14], and Baek

and Karniadakis [15]. By resorting to this robust method, it is expected that we

can go up to a very high Reynolds number case without unstable solutions.

Another reason for this discrepancy is attributed to the one dimensional

simplification to the experimental arrangement of multi-hinge connection. A

significant feature to be addressed further is the effect of the generator unit on

the VIV motion of the cylinder. These three aspects will be studied further.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 173

The present paper described some preliminary results of numerical simulation of

VIV phenomenon around a pivoted circular cylinder subject to steady flow. The

present flow model and method utilized an SEM-based N-S solver, which are

originally developed by Professor G.E. Karniadakis’ group. Numerical treatment

for movement of the body was done by using ALE approach. The solution

procedure and characteristic aspects of the present modeling and numerical

method were briefly stated. The present method of coupling of the body motion

with the flow problem is based upon a kind of weak coupling scheme. A series

of numerical estimation of VIV flow characteristics were carried out for a given

parameter set. By comparing with experimental results, it has been found out that

there is a great need to utilize a strong coupling scheme and to refine the

modeling of the multi-hinge connection of the system, which will be sought

sincerely for the forthcoming paper.

Acknowledgements

The present study is supported by the “Development of key-technology of ocean

renewable energy using Vortex-Induced Vibration (VIV) in water flow” granted

by Ministry of Knowledge Economy of Korea and the principal R&D program

of KORDI: “Performance Evaluation Technologies of Offshore Operability for

Transport and Installation of Deep-sea Offshore Structures” granted by Korea

Research Council of Public Science and Technology. The authors express deep

gratitude to Professor Karniadakis for permission of the computer code, known

as NEKTAR.

References

[1] Karniadakis, D., and S.J. Sherwin, Spectral/hp Element Methods for CFD,

Oxford University Press, 2nd Ed., London, 2005.

[2] Blevins, R. D. Flow-Induced Vibration, Malabar, FL Krieger Publishing

Company, 1994.

[3] Sarpkaya, T. A critical review of the intrinsic nature of vortex-induced

vibrations, Journal of Fluids and Structures, Vol. 19, pp. 389–447, 2004.

[4] Raghavan, K, “Energy Extraction from a Steady Flow Using Vortex

Induced Vibration”, Ph.D. Dissertation, The University of Michigan, 2007.

[5] Bernitsas, M.M. and Raghavan, K. “Reduction/Suppression of VIV of

Circular Cylinders through Roughness Distribution at 8×103< Re<1.5×105”,

Proc. OMAE Conference, pp. 1-5, 2008.

[6] Bernitsas, M.M., Raghavan, K., Ben-Simon, Y. and Garcia, E.M.H.

“VIVACE (Vortex Induced Vibration for Aquatic Clean Energy): A New

Concept in Generation of Clean and Renewable Energy from Fluid Flow”,

Proc. OMAE Conference, pp. 1-18, 2006.

[7] Ho, L.-W., and Patera, A.T., “A Legendre Spectral Element Method for

Simulation of Unsteady Incompressible Viscous Free-Surface Flows,”

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

174 Advances in Fluid Mechanics IX

1/3, pp. 355-366, 1990.

[8] Robertson, I., Sherwin, S.J., and Graham, J.M.R., “Comparison of Wall

Boundary Conditions for Numerical Viscous Free Surface Flow

Simulation,” Journal of Fluids and Structures, Vol. 19, No.4, pp. 525-542,

2004.

[9] Rabier, S., and Medale, M., “Computation of Free Surface Flows with a

Projection FEM in a Moving Mesh Framework,” Computer Methods in

Applied Mechanics and Engineering, Vol. 192, No. 41/42, pp. 4703-4721,

2003.

[10] Bouffanais, R., and Deville, M.O., “Mesh Update Techniques for Free-

Surface Flow Solvers Using Spectral Element Method,” Journal of

Scientific Computing, Vol. 27, No.1/3, pp.137-149, 2006.

[11] Papaioannou, G.V., D.K.P. Yue, M.S. Triantafyllou, G.E. Karniadakis, On

the effect of spacing on the vortex-induced vibrations of two tandem

cylinders, Journal of Fluids and Structures 24, pp. 833–854, 2008.

[12] Placzek A., Sigrist J.-F., Hamdouni A. Numerical simulation of an

oscillating cylinder in a cross-flow at low Reynolds number: Forced and

free oscillations, Computers & Fluids 38, pp. 80–100, 2009.

[13] Jong-Su Choi, Sup Hong, Tae-Kyeong Yeu and Hyung-Woo Kim, 2010,

“Experimental Study on Vortex-Induced Vibration Clean Energy Extraction

Device Model using Circular Cylinder Pivoted with a Lever (in Korean)”,

Proc. KSOE Spring conference, pp. 2516 – 2519, 2010.

[14] Baek, H. 2010, A spectral element method for fluid-structure interaction:

new algorithm and application to intracranial aneurysms. Ph.D. Thesis,

Brown University.

[15] Baek, H. and G.E. Karniadakis, 2010, Stabilization of fluid-structure

interactions via fictitious added mass and damping in physical and

biological applications, submitted to Elsevier.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 175

advanced materials used in aeronautics

O. Gohardani & D. W. Hammond

Department of Power and Propulsion, Cranfield University, UK

Abstract

In aircraft icing applications the interaction of Super-Cooled Liquid Droplets

(SLD) with the forward facing surfaces of an aircraft is of outmost importance,

as it influences the consequent icing that ensues on the structure. In this study,

empirical results are shown that characterize different droplet splashing scenarios

onto polymeric matrix composites reinforced with carbon nanotubes. Results are

presented for droplet interaction upon advanced pristine and eroded aerospace

materials based on a numerical scheme, that identifies different splashing features

into distinct categories. The purpose of utilizing the dynamic wetting analysis

scheme is to numerically represent the type of splashing events that are evident

on the experimental photographs. Images are presented for splashing features

on different surfaces and their similarities and differences are discussed based

on the analysis. Hydrophobicity and hydrophilicity of the different specimen

surfaces are further examined based on both static and dynamic contact angle

measurements and a correlation between the two modes of wetting is presented

for the aforementioned materials.

Keywords: dynamic wettability, contact angle, droplet interaction, carbon

nanotubes, aeronautics.

1 Introduction

and carbon nanotubes have widely been recognized due to the number of

possible applications where these materials may present desired characteristics. In

aerospace applications, composite materials are widely being considered because

of their higher stiffness and strength [1]. The present study investigates the

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120161

www.engbookspdf.com

176 Advances in Fluid Mechanics IX

occurs empirically. This interaction is comparable to the splashing that occurs upon

the airframe of an aircraft, flying through a cloud formation or when subjected to

precipitation. Weathering effects due to extended service operations of aircraft in

flight may occasionally result in structural wear on the airframe and erosion as

described in an extensive review by Gohardani [2]. For this purpose the considered

specimens have been modified by erosion, in order to investigate the role of surface

morphology on the ensuing splashing structures.

Wetting characteristics outside aerospace sciences have been examined in a wide

range of applications, where in particular the role of wetting on micro and nano-

structures in biological applications, has been studied for plant surfaces [3], insect

wing structures [4] and animal locomotion on the surface of water layers [5].

2 Experimental details

2.1 Experimental setup

The experiments in this study are carried out at the vertical droplet tunnel located

at Cranfield University, United Kingdom. The tunnel is located next to the main

Icing Tunnel, with the flow in both tunnels powered by a centrifugal backward

curved suction fan, capable of producing flow rates, ṁ ∈ [30, 100] kg·s−1 . The

temperature range T ∈ [−30, +5] ◦ C is attained by the refrigeration plant which

has a capacity of 400 kW. Upon generation of the flow, the air from the fan is

directed through a duct into a heat exchanger for cooling and directed into a

steering dish in the adjacent vertical tunnel where it gets accelerated toward the

test section. A droplet generator equipped with an interchangeable platinum nozzle

orifice disk is placed on top of the steering dish.

Upon entering the contraction section of the vertical droplet tunnel, the mono

dispersed droplets are accelerated towards the test section, where the gentle

contraction length of 5 meters, ensures that no aerodynamic breakup of the droplets

occurs. The test section is situated on top of two control valves, capable of

regulating the locus of the stagnation point. The target specimen is placed on a

designed target holder, that allows for vertical position adjustment of the specimen.

This is essential in order to account for the range of different specimen thicknesses

within the study. The incident angle of the oncoming air in relation to the target

is α = 70◦ . Upon contact with the target area the flow is bifurcated and exits the

tunnel through the control valves into the air inlet of the main Icing Tunnel, where

it gets regenerated by the fan. The droplet diameter in this study has been confined

to d = 400 µm, with three different free stream velocities U∞ ≈ {35, 50, 60}

m · s−1 , in accordance with expected number of splashing structures for statistical

purposes. The mono dispersed droplets are ejected from the orifice of the droplet

generator at a frequency of fd ≈ 12 kHz. Figure 1 shows the experimental setup

for the dynamic wettability experiments.

The illumination of the target area is attained by directing the light from an

LED through a collimating lens system. The LED device is strobed with a pulse

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 177

Figure 1: The experimental setup for the dynamic wettability experiments with

notations: (A) Vertical droplet tunnel, (B) droplet cloud, (C) CCD

camera, (D) laser sheet, (E) laser, (F) control valves, (G) LED strobing

device, (H) collimating lens, and (I) specimen target. The right hand

side image depicts a schematic corona structure with virtual planes

i0 , . . . , iN , with im showing the mid-plane.

strobe per acquired frame of the experiment. The image acquisition is actualized by

employing a Sony XCD-SX90 camera operated at 30 FPS. The camera is equipped

with a 50 mm lens and 13 mm extension. The distance between the camera and the

wall of tunnel dW = 50 mm, the LED and the collimating lens dC = 40 mm, and

further between the collimating lens and the wall of the tunnel is 100 mm.

The optical magnification of the experimental image is hence 0.32. The images

are acquired upon utilizing a graphical user interface on a personal computer.

As the LED light is captured along the length of the specimens, splashing features

are apparent in the volume of space along the width of each specimen. The

image acquisition camera is focused on the mid-plane of the specimen surface

as denoted by im in Figure 1. This entails that any structure appearing within

this plane of sight will appear entirely in focus on the experimental photographs.

Equally, a deviation from this plane will result in structures appearing not entirely

in focus. However, as the employed illumination within the depth of field is

adequate, even off-plane splashing features are distinguishable. This approach

hence results in capturing of most splashing events, at the expense of them

appearing simultaneously on similar locations along the width of the specimen

which may result in distinction of certain features becoming intricate. It is notable

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

178 Advances in Fluid Mechanics IX

that as the LED light travels through all virtual planes i0 , . . . , iN , an integrated

view is visible upon image acquisition.

In order to estimate the local liquid water content on the specimen surface, a

laser sheet with a 2 mm width was used to illuminate the exact half-width of

the specimen, where the camera was focused parallel to the oncoming droplets.

Upon entering the laser sheet, the droplets were illuminated and visible as distinct

features upon image acquisition. The finite thickness of the laser sheet further

implied that very few unfocused droplets were observed.

3 Experimental specimens

A total number of 8 different materials referred to hereafter as specimens

A−H were utilized in this study in two different prescribed conditions; as

supplied and in an eroded condition. The majority of these are established epoxy

resins commonly used within the aerospace industry with addition of carbon

nanotubes as a reinforcing agent. Specimens A and B are both combinations

of Araldite R

LY564/Aradur R

2954 with the exception that specimen B also

features multi-walled carbon nanotubes (MWCNT) at a 0.5% wt. Graphistrength R

R R

C100. Specimen C is Araldite DBF/Aradur HY956EN base resin with

10% wt. aluminium nitride nanoparticles. Specimens D−G are Araldite R

MY0510/Aradur 976-1 combinations where D and F have 0.5% wt. MWCNT.

R

considered material specimens have dimensions, (L × W ) = (30 × 20) mm2 with

different thickness values of H ∈ [1, 9] mm.

In the pristine state, no morphological treatment to the specimen surface is

applied. This condition replicates instances where no material degradation has

been observed upon implementation of the material on an aircraft during flight.

The second state is an eroded state of the material, which is attained upon its wet

blasting by using alumina. The choice of this abrasive stems from the need to

obtain a fine degradation of the material without catastrophic failure. An eroded

state of the material is obtained from an accelerated testing technique. However,

this approach allows for establishment of surface morphology modifications on

the resulting wetting characteristics of the materials. The surface roughness of the

specimens prior and after wet blasting is measured by a Surtronic instrument. The

average value of the surface roughness pre-erosion Rap , surface roughness post-

erosion Rae and the percentage change in surface roughness ∆Ra are shown in

Table 1.

3.1 Methodology

wetting characteristics, by employing the flowchart in Figure 2.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 179

results.

are acquired is chosen by the user. This velocity determines the local liquid water

content based on a numerical scheme that identifies the droplet cloud distribution

from experimental photographs and provides an averaged histogram of the cloud

spatial distribution on the target. Static and dynamic wetting photographs in the

pre- and post-erosion state of the specimen are further used as inputs to the

numerical specimen detection scheme. The resulting photographic information

is upon input, processed based on a number of Graphical User Interfaced tools

developed in M ATLAB R

, as shown in Figure 3. Upon post-processing of the data,

a decision matrix determines a probability factor which is attributed to a specific

specimen.

. The

tool outputs the projected area of the corona A, angles α and β, and

distances H, W1 , W2 and other image related properties.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

180 Advances in Fluid Mechanics IX

4 Data acquisition

4.1 Liquid Water Content

monitoring of the cloud scatter was actualized. The acquired photographs were

digitally processed based a numerical approach which subdivided each photograph

into 10 regions with the same size along the length of the specimen. The

photographic stack of 347 images, was after contrast enhancement, processed by

a droplet counting scheme which determined the number of droplets in each sub-

region. The averaged number of droplets within a sequence of images, in a given

zone were further given by

n

1

Rj = Di (1)

n i=1

j

where i = {1, 2, . . . , n}, j = {1, 2, . . . , 10} and n is the number of frames equal to

347. Upon known number of droplets, a similar approach for estimation of the local

Liquid Water Content (LW C) as Ide [6] was employed. For a constant droplet

diameter, d and total sample area for the bins A, the LW C can be expressed as

ρπd3

LW C ∼

= Rj (2)

∞t

6AU i

where ρ is the water density, t is the sampling time, U∞ is the free stream

velocity, and the sum refers to the total number of droplets. The local liquid water

content for free stream velocities U∞ ≈ {35, 50, 60} m · s−1 was estimated to

LW C ∼ = {0.38, 0.16, 0.02} g · m−3 . Figure 4 shows the averaged number of

droplets within 10 equally sized bins at three different free stream velocities for

a sampling duration of 10 seconds. The lack of droplets between the first and last

bin for free stream velocities above 35 m · s−1 indicate that lower local LW C

values are observed with an increasing U∞ . For statistical purposes U∞ ≈ 35

m · s−1 hence provides more reliable empirical results compared to the other free

stream velocities.

Both static and dynamic sessile drop methods were employed, in order to obtain

angles at which the droplets move along the surface of the specimen. These

methods convey results about the surface energy using Young’s equation

where γ is the surface energy and the subscripts SG and SL denote solid-gas and

solid-liquid respectively. In the static sessile drop method, the angles between the

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 181

a dark color for the first bin and a bright color for the last bin, for

different free stream velocities U∞ ≈ {35, 50, 60} m · s−1 along the

non-dimensional specimen length, .

sessile drop method, the drop is modified by either dispensing or retracting its

volume, resulting in an advancing angle θA or a receding angle θR , measured by

a protractor. This method allows for an assessment of the homogeneity of the

specimen, as droplets can be deposited on different locations on the specimen

surface. In this study an averaged value of five measurements placed on different

locations on the surface are presented. Hysteresis ψ, can in this context be defined

as ψ ≡ θA − θR . The equilibrium Young contact angle θC can also be expressed

in terms of the advancing and receding angles [8] as

ΓA cos(θA ) + ΓR cos (θR )

θC = arccos (4)

ΓA + ΓR

where ΓA,R ≡ {sin3 (θA,R )/(2 − 3 cos(θA,R ) + cos3 (θA,R ))}1/3 . Table 1, shows

the advancing, receding, equilibrium Young contact angle and their corresponding

hysteresis.

The static wetting of the specimens is assessed by employing contact angle

measurements on both pristine and eroded specimen surfaces. The experimental

setup for the contact angle measurements is shown in Figure 5. The specimen is

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

182 Advances in Fluid Mechanics IX

Table 1: Surface roughness on the specimens in the pre- and post-erosion state

given by Rap and Rae respectively. The percentage increase and decrease is

given by ∆Ra = (Rap −Rae )/Rap . The advancing angle θA , receding angle

θR , hysteresis ψ and the equilibrium Young angle θC for each specimen

are shown with the pristine value in the left and the eroded value to the

right in each column.

A 5 4.24 −0.152 110 75 22 11 88 64 22 11

B 0.74 4.4 4.946 93 90 33 6 60 84 35 6

C 0.8 3.74 4.946 80 90 17 10 64 80 18 10

D 0.2 3.98 3.675 87 80 28 10 58 70 30 10

E 1.2 3.92 2.267 93 70 32 8 62 62 33 8

F 0.2 4.06 19.30 90 68 28 13 62 55 30 14

G 0.8 2.32 1.90 82 110 32 22 50 88 34 22

H 0.6 2.98 3.967 117 90 60 8 57 82 63 8

light emitting diode with a diffuser, which provides a uniform illumination of the

droplet.

The setup is utilized for measurement of the advancing and receding contact

angles on the different specimen surfaces.

Figure 5: The experimental setup for the contact angle measurements, with

notations: (A) Eye piece fitted with a protractor, (B) specimen,

(C) generated droplet, (D) LED with diffuser, (E) vertical adjustable

table, and (F) syringe and spring device fitted with a micrometer.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 183

in m · s−1 .

U∞ Re We K Ca ξ∗

35 12 271 6 725 49 856 0.55 3.03 − 3.04

50 17 530 13 725 101 747 0.78 3.32

60 21 035 19 764 146 567 0.94 3.47

The splashing phenomenon is often related to different non-dimensional

parameters such as the Reynolds number Re ≡ (ρU √∞ d)/µ, the Weber number

2

We ≡ (ρU∞ d)/σ, the Ohnsorge number Oh ≡ We/Re, and the Capillary

number Ca ≡ We/Re, where d is the droplet diameter, ρ defines the density, U∞

is the free stream velocity, µ is the dynamic viscosity and σ denotes the surface

tension. The spreading factor ξ ∗ ≡ dmax /d, can further be expressed as [7]

We + 12

∗

ξ =

(5)

4We

3 1 − cos(θA ) + √

Re

Table 2, shows the non-dimensional parameters for the considered free stream

velocities.

It is notable that the influence of the advancing angle on the spreading factor is

nominal for the considered cases as it influences the maximum spread diameter,

dmax with less than 5%. From Table 2, it is evident that dmax ∈ [3.0d, 3.5d]

where d, is the initial diameter of the droplet. Representative photographs for each

specimen in its pristine and eroded state respectively, are shown in Figure 6.

5 Discussion

The general observation for the wetting characteristics of the considered

materials exhibits that specimens in the pre-eroded state are mainly hydro-

phobic as indicated by the formation of discrete water beads upon wetting. With

modifications to the surface morphology by erosion, a partial or complete water

film is apparent on all specimens.

The difference between the specimens in the post-eroded state is therefore,

solely determined by the number of streams along the specimen surface. For

the specimens exhibiting completely continuous water films, these streams merge

to one coherent water film flowing along the specimen surface. By contrast,

specimens with partial wetting characteristics have a discrete number of streams

separated by non-wetted regions between them.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

184 Advances in Fluid Mechanics IX

A−H. The brightness and contrast of the images have digitally been

enhanced for visualization purposes.

6 Conclusions

characteristics in the pristine state of each material, with the materials being

largely hydrophobic. Upon erosion, the formation of water films on the specimen

surfaces, indicate a more hydrophilic behavior. Despite the complexities of

image processing for different static and dynamic wettability arrangements, a

methodology has been developed that discriminates between different empirical

photographs of hydrophobic and hydrophilic specimens based on a numerical

scheme. This scheme has thus far provided satisfactory results based on post-

processing of the empirical photographs. For more adequate results however

new modules have to be added to this methodology that would allow for better

identification rules of different materials, based on their wetting characteristics.

Acknowledgement

The authors greatly acknowledge the financial support from the European Union

for conducting this research as a part of multi-functional layers for safer aircraft

composite structures (LAYSA) project, under the Seventh Framework Programme

(FP7), including aeronautics.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 185

References

[1] Shalin RE. Polymer matrix composites. 1995; Springer, pp 440. ISBN

0412613301.

[2] Gohardani O. Impact of erosion testing aspects on current and future flight

conditions. 2011; Progress in Aerospace Sciences. Volume 47, Issue 4,

Elsevier, pp. 280–303. doi:10.1016/j.paerosci.2011.04.001.

[3] Neinhuis C and Barthlott W. Characterization and distribution of water-

repellent, self-cleaning plant surfaces. 1997; Ann Bot, 79(6): pp 667–677.

[4] Byun D, Hong J, Saputra, Ko JH, Lee YJ, Park HC, Byun BK, Lukes JR.

2009; Wetting Characteristics of Insect Wing Surfaces. J Bionic Eng, Vol

6(1), pp 63–70.

[5] Hu DL, Prakash M, Chan B, and Bush JWM. Water-walking devices.

In; Animal locomotion. 2010; Springer. Eds: Taylor G, Triantafyllou MS,

Tropea C, pp 350. ISBN 3642116329.

[6] Ide RF. Comparison of Liquid Water Content measurement techniques in an

icing wind tunnel. 1999; NASA/TM–1999-209643.

[7] Pasandideh-Fard M, Qiao YM, Chandra S, and Mostaghimi J. Capillary

effects during droplet impact on a solid surface. 1996; Phys Fluids 8(3), pp

650–659.

[8] Tadmor R. Line Energy and the Relation between Advancing, Receding, and

Young Contact Angles. 2004; Langmuir 20(18), pp 7659-7664.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

This page intentionally left blank

www.engbookspdf.com

Section 3

Multiphase flow

www.engbookspdf.com

This page intentionally left blank

www.engbookspdf.com

Advances in Fluid Mechanics IX 189

humidity influence on evaporating water

droplets and vapour condensation intensity

G. Miliauskas, S. Sinkunas & K. Sinkunas

Department of Thermal and Nuclear Energy,

Kaunas University of Technology, Lithuania

Abstract

This paper discusses modelling of water droplet heating and evaporation in

humid air. The effect of black body spectral radiation of the air temperature on

the combined heat transfer in the water droplet has been evaluated. A combined

analytical and numerical method of investigation is applied. Securing the balance

of energy fluxes in the droplet with a confidence of one hundredth of a percent,

the method of fastest descent has been used to determine the droplet surface

temperature. Thermal state and phase transformation dynamics for the warming

droplets heated in a humid air environment are modelled under conductive and

combined conductive-radiative heating.

Keywords: humid air, water droplet, combined heat and mass transfer, droplet

heating and evaporation, condensation.

1 Introduction

The research of water droplet heat and mass transfer is connected with

application of sprayed water technology in power energy and other industries.

The problem of droplet evaporation has been researched for more than a hundred

years [1]. Demand to improve technology which employs liquids in the sprayed

form requires further understanding of combined heat and mass transfer

processes in liquid droplets under wide range of conditions. Applied methods for

investigation of heat and mass transfer in liquid droplet are systematically

reviewed in [2]. There has been an objective for particular consideration of

unsteady combined heat and mass transfer processes in recent years [3–9]. The

methods developed for analytical and numerical investigation of droplet

combined heat and mass transfer allows modelling of more complex processes

and their interactions.

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120171

www.engbookspdf.com

190 Advances in Fluid Mechanics IX

specifically to durability of boiler tubes. Hence, understanding of factors

affecting tube metal degradation is significant to nuclear safety of the plant. The

relevant degradation mechanism depends on tube wall metal and the

environment properties it operates in. The risk of stress corrosion cracking is

applicable to stainless steel tubes, which normally operate with superheated

steam, but under certain operational conditions have likelihood of droplet

carryover into that section of the boiler.

Water droplet heating and evaporation can be modelled in different ways.

Various aspects of heat and mass transfer can be considered: droplet slipping in

steam flow will affect convection, droplets in steam are also heated by

convection, but radiation of the surroundings determines if heating is combined

heating. Sprayed droplet initial temperature and dispersity are also very

important factors affecting heat and mass transfer. The condensation process

requires the presence of vapour in the droplet surroundings. The initial droplet

temperature affects if heat and mass transfer on the droplet surface will be for

condensation or evaporation. For small droplets of diameter of microns range it

is necessary to evaluate the influence of Knudsen layer to droplet evaporation. In

the case when droplets are large the influence of surroundings radiation must be

evaluated. Each of these factors contributes essential peculiarities for heat and

mass transfer modelling [9–11].

The results of analysis of heating and phase transformation for larger droplets

on their surface in humid air are presented in this paper.

The intensity of phase transformations at the droplet surface determines water

vapour flux [12]:

Dvg v p pv,

mv

pv, R pv, v p ln

p p

pv, pv, R .

(1)

Tvg , R R R g v, R

negligibly small, and steam and gas mixture temperature near the droplet in

expression (2) is comparable to the droplet surface temperature: Tvg , R TR . The

droplet surface temperature determines the regime of phase transformations –

when it is below the dew point temperature the water vapour of humid air

condenses on the droplet surface. When heated droplet surface temperature

reaches dew point temperature the regime of phase transformation changes and

droplet starts to evaporate. Air humidity is defined by volumetric part of water

vapour pv pv , / p . The driving force of diffusive evaporation is defined by the

water vapour pressure difference near the droplet and in humid air. The effect of

Stefan hydrodynamic flow on the intensity of evaporation is evaluated by

logarithm in expression (1). The droplet surface temperature is determined by

heat flow interaction at the surface. This temperature must secure the heat flux

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 191

in-flow and out-flow balance. Assumption of heat and mass transfer process

quasi-stationarity enables to describe the heat flow balance by the condition:

vg 1 BT T r ,

qk qk q f 0;

R

Tg TR ln BT

r r R

mv L . (2)

takes into account effect of Stefan hydrodynamic flow on the intensity of

convective heating of evaporating droplet. Water circulation in the droplet has

not been considered. It is assumed that there is no phase slippage in humid air

flow carrying water droplets. The incident spectral radiation onto droplet is

partially reflected and partially absorbed in the semitransparent droplet. The

direction of vapour flux is determined by the logarithmic function in expression

(1), and it is considered being positive in the case of evaporative regime. The

temperature gradient for combined conductive-radiative heat transfer process in

the droplet is described by expression [10]:

T r , 2

n 1 1

R dTR

2

n n

r r R R n 1 0 n d *

(3)

R n 2

1

sin n n cos n q dr exp a * d * .

Rc p

r

0 R

Flux density of radiation in spherical semitransparent droplet is calculated

according to methodology [10] and the complex refractive index of water is used

in accordance with [13, 14] recommendations. The Spalding heat transfer

number takes into account the radiation absorption by droplet and temperature

gradient in it [15]. The expression (2) is solved numerically. The time step is

chosen freely. Instant droplet surface temperature is determined by the iterative

method of fastest descent. Imbalance of heat fluxes on the droplet surface is

achieved to less than one hundredth of a percent. After calculating the vapour

flux on droplet surface in accordance with expression (1) the volume change of

spherical droplet in time step Δτ is determined by:

V

4R 2 mv . (4)

To determine the droplet surface temperature its volume is assumed constant

during time step Δτ while conducting iterative calculations. This ensures the

stability of the numerical iterative scheme. The droplet volume is revised for

every next time step.

The analysis has been conducted for water droplets of various initial

temperatures in the humid air (of pv 0.2 0.95 ) of 700 K and 0.1MPa.

Conductive heating has been modelled assuming that droplets in air are stagnant.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

192 Advances in Fluid Mechanics IX

In the case of combined heating by conduction and radiation the external black

body radiative source with air temperature has been assumed. Droplet

evaporation process is significantly influenced by the degree of moisture in the

air (Fig. 1). Curves describing the change of heat and mass transfer parameters

presented in the dimensionless form of Fourier number scale are general for all

conductively heated droplets. The initial water temperature and air parameters

must be defined.

R

R3

1.25 R

R03

1

1 2

3

0.75

1

4

2

0.5 3

4

300K

0.25 a)

370K

0

Fo

0 4 8 12 16

pv, : (1) 0.2, (2) 0.4, (3) 0.6, (4) 0.8.

humidity and sprayed water temperature. Volume of the low initial temperature

droplets increases during the primary stage of phase transformations due to

vapour condensation on their surfaces and expansion of warming water. The

droplet volume is increasing for certain duration due to expansion of warming

water during the initial stage of the droplet evaporation regime. The evaporating

droplet volume starts to decrease when the process of water evaporation exceeds

water expansion process. The effects of above mentioned competing processes to

the dynamics of droplet volume reach equality and are visible on the volume

dynamics curve R Fo as an extremum point (Fig. 1). The droplet volume of

high initial water temperature does not have volume expansion stage and its

volume starts to decrease immediately. Rapid volume decrease is caused by

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 193

quite quickly reach equilibrium evaporation regime. Equilibrium evaporation is

defined as a phase transformation regime during which heat delivered to the

droplet by the surroundings is solely consumed for droplet evaporation, i.e.

q f Fo Foe q . The beginning of equilibrium evaporation for conductively

heated droplets is also indicated as convergence of curves q f Fo and qk Fo .

kW

q,

m2

1

200

2

4

150

100

1 q q k

5

6

2 q q f

50

T0=300K

0

0 4 8 12 Fo16

Figure 2: Heat fluxes on the droplet surface for conductively heated cold

droplets. pv, : (3) 0.2, (4) 0.4, (5) 0.6, (6) 0.8; R0 0.0001m.

The temperature of the conductively heated droplets does not change during

the equilibrium evaporation regime (Fig. 3). The equilibrium evaporation

temperature is reached by intensive droplet heating for low initial temperature

droplets (Fig. 3a) and intensive cooling for high initial temperature droplets

(Fig. 3b). The distinct non-isothermality is observed during the initial stage of

phase transformations. The droplet non-isothermality diminishes when

approaching the regime of equilibrium evaporation. The near equilibrium

temperature is first reached by the subsurface layers in the droplet and later by

the central layers of the droplet.

During the unsteady phase transformations not only the droplet temperature

changes rapidly, but also the heat fluxes at the droplet surface change

significantly (Fig. 4). This is caused by the initial water droplet temperature and

their way of heating.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

194 Advances in Fluid Mechanics IX

T, K

360 4

3

350

2

340

1

330

320 TR Tm TC

310

T0=300K a)

300

0 0.25 0.5 0.75 1 1.25

Fo

T, K

1 2 3

370

4

360

350

T0=370K

b)

340

0 0.25 0.5 0.75 1 Fo

temperature dynamics for conductively heated droplets during the

initial stage of phase transformations. pv, : (1) 0.2, (2) 0.4, (3)

0.6, (4) 0.8.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 195

kW

q,

m2 k

150 k+r

T0=300K

125

1

100

1 q q k

75 2

2 q q f

50 3 q q k

4 q qr

25 3

4 a)

0

0 0.25 0.5 0.75 1 1.25 Fo

kW

q,

m2 k

k+r

250

1 q q k

200 2 q q f

2 3 q q k

150 4 q qr

1

100

T0=370K

3

50

4 b)

0

0 0.2 0.4 0.6 0.8 Fo

Figure 4: Heat fluxes at the droplet surface for low and high temperature

droplets during unsteady phase transformation regime for the

conductive heating case (k) and the combined heating case (k+r).

pv, : (1) 0.2, (2) 0.4, (3) 0.6, (4) 0.8.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

196 Advances in Fluid Mechanics IX

significant (Fig. 5). Intensive heating of low initial temperature droplets up to

dew point temperature is caused by intensive vapour condensation process

(Fig. 4a). When the droplet is heated conductively, then the intensity of water

warming is determined by conductive heat flux on the internal side of the droplet

surface, and in the case of combined heating – by the total heat flux. The total

heat flux on the internal side of the droplet surface during the regime of

condensational phase transformation is q 0 Foco qk qr q f co , where

term qco approaches zero. At the moment of phase transformation regime

changeover the condition qk q k is valid. Accelerating process of evaporation

inhibits water warming in the droplet: qh Foco Foe qk qr q f . The phase

transformation heat flux during the unsteady evaporation regime is

q f Foco Foe q q . The dynamics of heat fluxes during the unsteady

evaporation regime determines the initial spayed water temperature and the ways

droplet are heated (Fig. 4). The influence of heating manner is especially distinct

for low temperature droplets (Fig. 4a). Conditions for absorbed radiation in the

semitransparent droplet to participate in the water evaporation process establish

only when droplet temperature field of negative gradient is formed. The

conductive heat flux becomes zero at the moment of the vector direction

mv

mv, 0 k+r

1

pv , 0.6

0.8

k

1 2

0.6 3 4

5

0.4

pv , 0.2

k+r

0.2

k

T0=300K

0

0 1 2 3 4 Fo

(2) 0.00005, (3) 0.0001, (4) 0.0002, (5) 0.0003.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 197

(Fig. 4a). The condensational phase transformation regime for the high initial

water temperature droplets is impossible. The temperature field of negative

gradient is formed immediately at the start of evaporation. The rapid cooling of

the droplet is caused by very intensive evaporation process (Fig. 4b), which is

associated with the participation of cooling droplet internal energy. The droplet

cooling is slowing with approach to the equilibrium evaporation regime and the

influence of the droplet internal energy reduces. For equilibrium evaporation

regime the gradient of temperature field in the droplet ensures participation of

the droplet absorbed radiation flux in the evaporation process.

4 Conclusions

The regime of unsteady phase transformations for water droplets in humid air is

short and droplets rapidly reach the characteristic equilibrium evaporation

temperature. Nonetheless, the regime of unsteady phase transformations for

interaction of heat and mass transfer processes in sprayed water systems is very

significant. The way of droplet heating and water initial temperature play an

important role during this phase transformation regime.

The process of water vapour condensation, which is contained in air, causes

rapid warming of low initial temperature droplets to the equilibrium evaporation

temperature. The participation of droplet internal energy in the evaporation

process determines rapid cooling of high initial temperature droplets. The initial

water droplet temperature does not influence heat transfer processes during the

equilibrium evaporation regime.

The effect of radiation to temperature dynamics and phase transformations is

more significant for droplets of bigger size than for smaller ones. The influence

of radiation is very significant for conductive heat flux dynamics in the cold

water droplets.

5 Nomenclature

a – thermal diffusivity, m2/s; BT – Spalding heat transfer number; cp – specific

heat, J/(kg K); D – mass diffusivity, m2/s; Fo – Fourier number; k – conductive

heating; k+r combined heating; L – latent heat of evaporation, J/kg; m – vapour

mass flux, kg/(s.m2); n – number of terms in the infinite sum; p – pressure, Pa;

q – heat flux, W/m2; R – radius of a droplet, m; Rμ – universal gas constant,

kg/(kmol K); r – radial coordinate, m; T – temperature, K; V – droplet volume,

m3; η – non-dimensional coordinate; λ – thermal conductivity, W/(m K); μ –

molecular mass, kg/kmol; ρ– density, kg/m3; τ – time, s;

Subscripts: C – droplet centre; co – condensation; e – equilibrium evaporation;

f – phase transformation; g – gas; k – conductive; m – mass average; r –

radiative; R – droplet surface; rt – dew point; v – vapour; vg – vapour-gas

mixture; 0 – initial state; – far from a droplet; Σ – total.

Superscripts: + – external side of the surface; - – internal side of the surface.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

198 Advances in Fluid Mechanics IX

References

[1] N.A. Fuks. Evaporation and Droplet Growth in Gaseous Media. London:

Pergamum Press, 1959.

[2] Sazhin S.S., 2006, Advanced models of fuel droplet heating and

evaporation, Progress in Energy and Combustion Science, Vol. 32, pp. 162-

214.

[3] Miliauskas G., Sinkūnas S. and Norvaisiene K., 2011, Influence of thermal

radiation to the interaction of heat and mass transfer processes under

evaporation of water droplets. ICCHMT'2011: 7th Inter-national

Conference on Computational Heat and Mass Transfer. Turkey: Istanbul,

pp. 1–8.

[4] Dombrovsky L.A., Solovjov V.P. and Webb B.W., 2011, Attenuation of

solar radiation by a water mist from the ultraviolet to the infrared range. J.

Quant Spectroscopy Radiative Transfer, Vol. 112, pp. 1182–1190.

[5] William F., Godoy and Paul E., 2009, Radiation driven evaporation for

polydisperse water sprays. Int. J. of Heat and Mass Transfer, Vol. 52,

pp. 2893–2901.

[6] Miliauskas G., Sinkūnas S. and Miliauskas G., 2010, Evaporation and

condensing augmentation of water droplets in flue gas. Int. J. of Heat and

Mass Transfer, Vol. 53, pp. 1220–1230.

[7] Arias P.G., Im H.G., Narayanan P. and Trouve A., 2011, A computational

study of non-premixed flame extinction by water spray. Proceedings of the

Combustion Institute, Vol. 33, pp. 2591–2597.

[8] Hitoshi Fujimoto, Yosuke Oku, Tomohiro Ogihara and Hirohiko Takuda,

2010, Hydrodynamics and boiling phenomena of water droplets impinging

on hot solid. International Journal of Multiphase Flow, Vol. 36, pp. 620–

642.

[9] Tseng C. C. and Viskanta R., 2006, Enhancement of water droplet

evaporation by radiation absorption. Fire Safety J., Vol. 41, pp. 236-247.

[10] Miliauskas G., 2001, Regularities of unsteady radiative-conductive heat

transfer in evaporating semitransparent liquid droplets. Int. J. of Heat and

Mass Transfer, Vol. 44, pp. 785-798.

[11] Miliauskas G. and Garmus V., 2009, The peculiarities of hot liquid droplets

heating and evaporation. Int. J. of Heat and Mass Transfer, Vol. 52, pp.

3726–3737.

[12] Kuzikovskij A.V., 1970, Dynamic of spherical particle in powerful optical

field. Izv. VUZ Fizika, Vol. 5, pp. 89–94.

[13] Hale G.M. and Querry M.R., 1973, Optical constants of water in the 200-

nm to 200-m wavelength region. Appl. Opt., Vol. 12, pp. 555–562.

[14] Hale GM, Querry MR, Rusk AN and Williams D 1972 Influence of

temperature on the spectrum of water. J. Opt. Soc. Am., Vol. 62, pp. 1103–

1108.

[15] Miliauskas G and Sabanas V 2006 Interaction of transfer processes during

unsteady evaporation of water droplets Int. J. Heat Mass Transfer 49 1790–

1803.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 199

under enhanced gravity studied with

in-situ microscopy

T. Krebs1,2 , J. J. Slot3 , C. P. G. H. Schroen2 ,

H. W. M. Hoeijmakers3 & R. M. Boom2

1

Institute for Sustainable Process Technology, The Netherlands

2

Wageningen University, Food Process Engineering Group,

The Netherlands

3

University of Twente, Engineering Fluid Dynamics Group,

The Netherlands

Abstract

We report the results of experiments and numerical calculations of compression

and coalescence in a monodisperse oil-in-water emulsion upon centrifugation.

A custom-built setup allows in-situ monitoring of a rotating bilayer of emulsion

droplets using an optical microscope. The oil volume fraction in a compressed

layer of oil droplets stabilized against coalescence was measured experimentally

as a function of time for different radial accelerations. The sedimentation was

simulated using CFD in order to test the applicability of the computational method

and the Ishii-Zuber drag law for very high dispersed phase volume fractions.

Quantitative agreement of emulsion sedimentation as a function of time between

the experiments and simulations is good at higher accelerations, but decreases

with decreasing accelerations. Coalescence in a centrifuged emulsion, which

was destabilized prior to centrifugation by adding sodium chloride, was also

quantified. The growth of a pure oil phase on top of the droplet layer was measured

as a function of time. From the growth rate, a characteristic time for droplet

coalescence with the pure oil phase was deduced. The experimental method may

serve as a tool to study the compression and coalescence kinetics of emulsions

under enhanced gravity, which may be of use to assess emulsion stability for

industrial applications. Possible improvements of the current experimental setup

are also discussed.

Keywords: emulsion, compression, coalescence, microscopy, droplet, CFD,

drag law.

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120181

www.engbookspdf.com

200 Advances in Fluid Mechanics IX

1 Introduction

The behavior of dense emulsions in flow fields with radial accelerations is of major

importance for separation processes using cyclonic devices or centrifuges. Major

examples are separation of crude oil and water during oil production, and milk

processing in the food industry. The design of these separation processes is often

strongly based on the results of Computational Fluid Dynamics, which has become

an important tool for the process industry in recent years. For reliable modeling,

information about the sedimentation rate, and kinetics of coalescence and breakup

of droplets in a dense emulsion under enhanced gravity is needed.

A key parameter for modeling is the choice of the drag law for droplets as it will

impact the calculated rate of sedimentation. For dense droplet-laden flows, the drag

law of Ishii and Zuber is often used [1]. The Ishii-Zuber drag law is an extension

of the well known Schiller-Naumann drag law [2]. It can be applied to flows with

higher volume fractions of the disperse phase by incorporating hindered settling.

As a droplet moves through the fluid it will induce a motion of the continuous

phase and thereby deform the surrounding fluid. When other droplets are present

in this surrounding fluid they will be subjected to this deformation as well. Due the

Laplace pressure, the surrounding droplets will resist deformation more than the

continuous fluid, leading to a higher local viscosity acting on the moving droplet.

Therefore, the drag for these dense systems is modeled by assuming similarity to

the single droplet case and using an expression for the mixture viscosity instead of

the continuous phase viscosity. Besides the Ishii-Zuber drag law, other drag laws

have been proposed for dense systems [3]. Still, the Ishii-Zuber law is a popular

choice for CFD since it can be applied for a wide range of multiphase systems and

flow parameters.

In addition to applying proper relation for the drag law, a correct prediction

of the droplet size distribution and its effects on the flow field will increase

the accuracy of multiphase flow CFD. In multiphase CFD, droplet breakup and

coalescence kinetics can be taken into account using the approach of population

balance equations (PBE) [4]. Analytical hydrodynamic models to predict the

timescale of coalescence between two droplets have been proposed, which cannot

generally be applied to dense flowing emulsions under enhanced gravity, however.

Apart from analytical models, input for the coalescence functions in PBE modeling

can also be provided by experiments, in principle. For oil/water separation

applications, the kinetics of coalescence between droplets will impact the rate of

separation. The mean droplet size will increase in the system when coalescence

dominates over breakup, which will then accelerate the separation.

The dynamics of coalescence in centrifuged emulsions have been recently

investigated by us using a tabletop centrifuge [5]. From the experiments, we

derived characteristic coalescence times of the droplets in the emulsion as a

function of the radial acceleration and surfactant concentration. Analysis of the

samples was performed ex-situ, however, which increased the experimental error

and made the experiments time-consuming.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 201

under enhanced gravity using a custom-built centrifuge, which allows in-situ

microscopic observation of the emulsion during centrifugation. The degree of

compression in a stable emulsion is expressed as the oil volume fraction in

the compressed droplet layer. We compare the experimental results with CFD

calculations to test the applicability of the computational method and the Ishii-

Zuber drag law to high dispersed phase fractions. Further, we measure the rate of

coalescence in an unstable emulsion as a function of the radial acceleration. From

the coalescence experiments, characteristic times for coalescence of a droplet with

a pure oil phase under enhanced gravity can be derived.

2.1 Chemicals

Sodium n-dodecyl sulfate (SDS, ACS reagent, ≥ 99%) and sodium chloride

(NaCl, ACS reagent, ≥ 99%) were purchased from Sigma Aldrich. Sylgard 184

Silicone Elastomer kit was purchased from Dow Corning. As dispersed phase

Sil180 silicone oil (Dow Corning) was used. At 293 K, the density of Sil180 is

931 kg m−3 and the viscosity 10.4 mPa s. Aqueous solutions of NaCl (5 wt%)

and SDS (10 mM) were prepared with Millipore water. The density of the 5 wt%

aqueous NaCl solution was 1032 kg m−3 at 293 K. The density difference between

pure water and the oil is ∆ρ = 69 kg m−3 , and for the salt solution and the oil

∆ρ = 101 kg m−3 .

displays a photo of the experimental setup. The setup consists of a disk, which

is connected to a DC motor. The range of accessible rotation frequencies f is

1.5 Hz < f < 100 Hz. A sample holder can be mounted on the disk. The distance

r between the center of the sample holder and the motor axis is 10 cm. The

radial acceleration a is given by a = (2π f )2 r. The range of accelerations is thus

3.95 m s−2 < a < 39500 m s−2 , or 0.402 < a/g < 4020, if expressed in multiples

a/g of the normal gravitational acceleration g = 9.81 m s−2 . For the entire range

of a/g the induction time of the motor to reach 95% of the desired acceleration

was < 0.5 s. The disk and the motor are enclosed in a protective housing. The

entire assembly is placed on a custom-made plate, which is mounted on an inverted

optical microscope (Axiovert 200, Zeiss). Three set screws allow to align the

sample with the objective of the microscope. A high-speed camera (Y4-S2, IDT

Inc.) is connected to the microscope. A software trigger was used (Motion Pro

Studio, IDT Inc), which triggered recording when at least one pixel in the image

was not black, indicating that the sample chamber, which is transparent, is passing

by the microscope. A frame rate of 5000 Hz and an exposure time of 100 ns were

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

202 Advances in Fluid Mechanics IX

the sample chamber used for the centrifuge experiments. The chamber

is made from two PDMS disks. The bottom disk contains a channel

of dimensions 5.25 × 1 × 0.2 mm−3 , where the emulsion is inserted.

Panel C: photograph of a oil-in-water emulsion in the sample chamber.

The oil phase was Sil180 silicone oil, the aqueous phase a 10 mM SDS

solution. The top layer of a bilayer of monodisperse droplets with a mean

diameter 79.3 µm is displayed. The position of the second droplet layer

is indicated by the dotted circles.

used for the experiments. The recorded images were processed with the program

ImageJ using custom-written scripts.

A drawing of the sample chamber, which is placed in the sample holder, is

displayed in Figure 1B. The sample chamber consists of two parts, which are

made from polydimethylsiloxane (PDMS). The bottom part contains a channel

(dimensions 5.25 × 1 × 0.2 mm3 ), which is open to one side. The top part of the

sample holder contains no features. The parts are made from mold casting using the

Sylgard Elastomer Kit, which consists of liquid PDMS and a curing agent [6]. The

top and bottom part are chemically bonded together after exposure to an oxygen

plasma. Before each experiment, the sample chamber was exposed to an oxygen

plasma to render the chamber walls hydrophilic, in order to enable wetting of the

channel walls with the continuous phase [6].

Monodisperse droplets of Sil180 silicone oil in a 10 mM aqueous SDS

solution were produced with a microfluidic T-junction. The experimental setup

and parameters have been described previously [7]. An oil-in-water emulsion with

a mean droplet diameter d = (79.3 ± 0.8) µm was obtained. The emulsion was

stable for the entire duration (∼ 4 weeks) of the experimental work. The emulsion

creamed in the storage flask; droplets were removed from the dense layer using

a pipette and injected into the sample chamber. The mean oil volume fraction φ0

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 203

counting the number of droplets after spreading of the test volume on a microscope

slide. From ten measurements φ0 = 0.72 ± 0.03 was obtained. Figure 1C displays

a photograph of the emulsion in the sample chamber. The droplets have a circular

cross section and form a bilayer. In the image shown, the image was focussed

on the top layer of droplets. The second layer of droplets is visible only diffuse,

the position of the droplets relative to the first is sketched by the dotted circles in

Figure 1C.

24

CD = (1 + 0.15Re0.687

m ) (1)

Rem

The Reynolds number based on the mixture viscosity is defined as

Rem = ρw | uo − uw | d/ηm . Finally, the mixture viscosity is given by

ηo +0.4ηw

ηm = ηw (1 − φ )−2.5 ηo +ηw (2)

Here ηo and ηw are the dynamic viscosities of the oil and water respectively.

The governing equations were solved using the commercial CFD package Ansys

CFX 13.0. A mesh of 3.3 million hexahedral elements and the use of double

precision executables are required to capture the steep gradients in volume fraction

and to avoid strong oscillations near that gradient in the numerical solution. A no

slip condition is imposed on the walls. The use of a spatially varying centrifugal

force lead to numerical oscillations in volume fraction. Therefore, a constant body

force is used here. Due the large radius at which the chamber is placed compared

to its dimensions of the error made here is smaller than 1%. All other parameters

used in the calculation, such as d, a and φ0 are matched with the experimental

values. As initial condition a homogeneous emulsion with φ0 is used.

3.1 Emulsion compression

Figure 2A displays a contour plot of the oil volume fraction φ in the sample

chamber during centrifugation obtained from the CFD calculations at 103g after

10 s of centrifugation as an example. Due to the density difference between

the two phases (∆ρ = 69 kg m−3 ), oil is concentrated at the right side of the

sample chamber. At the left side of the sample chamber, pure continuous phase

is accumulated. The transition between the oil-rich and oil-free layer is sharp, as

can be seen from a cross-sectional plot of φ along the y-axis (Figure 2B). In the

oil-rich layer, φ (y) increases with increasing distance from the boundary between

the two layers, and eventually reaches unity. From the φ (y) profiles we calculate

an average oil volume fraction φc of the oil-rich layer, in order to compare the

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

204 Advances in Fluid Mechanics IX

of the oil volume fraction in the sample chamber during centrifugation

at 103g, 10 s after the start of centrifugation. Panel B: cross-sectional

profile of the oil volume fraction along the y-axis at x = 0.5 mm,

extracted from the contour plot in Panel A.

CFD results with the experiments. For the example shown in Figure 2B, φc = 0.87

is obtained.

Two images from the compression experiments are shown in Figure 3 as

an example. Panel A displays the emulsion at rest before centrifugation. The

mean volume fraction of oil is φ ≈ 0.72 . Panel B displays a snapshot of the

emulsion during centrifugation at 231g, 5 s after the start of centrifugation.

Upon centrifugation, droplets are accelerated towards the axis of rotation, thereby

forming a dense layer. From the height of the compressed droplet layer hc and

the total height of the liquid column ht we can calculate the averaged oil volume

fraction φc in the compressed layer, φc = φ0 ht /hc . For the example shown in

Figure 3B, we obtain φc ≈ 0.83. Figure 3 displays φc as a function of time, obtained

from the experiments and CFD calculations, for three different radial accelerations.

The experimentally measured φc (t) increase with time, but eventually approach

asymptotic values. For a given time, φc increases with increasing acceleration.

The maximum volume fraction for hard spheres is ≈ 0.74, which is close to

the experimentally measured value of the monodisperse dense droplet layer from

where the emulsion was sampled. For droplets, higher packing fractions can be

achieved, as droplets are deformable. The Bond number Bo = ∆ρ gd 2/σ compares

the magnitude of body forces and interfacial tension forces. For our system, at the

smallest acceleration of 103g, we obtain Bo ≈ 0.11, which indicates that droplet

deformation will take place [9], and that the measured values of φc > 0.72 are

physically reasonable. The maximum oil volume fractions at the corresponding Bo

are: φc,max (Bo = 0.11) = 0.75, φc.max (Bo = 0.25) = 0.79 and φc,max (Bo = 0.55)

= 0.90. The φc (t) curves obtained from the CFD calculation also increase with

increasing t, and φc (t) also increases with acceleration for a given t. The curves

do not approach an asymptotic value, however, and for 103g and 231g also differ

quantitatively from the experimental data.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 205

is 79.3 µm, the oil volume fraction is 0.72. Panel B: Photograph of

the same emulsion as in panel A after 5 s of centrifugation at 231g.

hc and ht indicate the height of the compressed emulsion layer and

total height of the liquid column, respectively. Panel C: mean oil

volume fraction φc in the compressed droplet layer as a function of

time for different accelerations. Filled symbols are the results from

microcentrifuge experiments. Each data point is the average of 4

individual measurements. The error bars are 95% confidence intervals.

Open symbols are the results from the CFD calculations.

The agreement between the φc (t) curves which were experimentally measured

and calculated from CFD is not good, both from a quantitative and qualitative point

of view. Generally, the calculated values overestimate the experimental values.

One reason for the discrepancy is the fact that in the computational model the

oil volume fraction will eventually approach unity for any radial acceleration,

as the dispersed phase is not treated as discrete particles, but as a continuous

medium. This is a direct consequence of the averaging procedure carried out in

the derivation of the two-fluid model. In the experiments, on the other hand, there

will be a maximum value of the oil volume fraction for a given Bo, for each radial

acceleration. The applicability of the computational method for high dispersed

phase fractions may be enhanced by imposing a maximum dispersed phase fraction

φmax that can be reached in a given volume element for the CFD calculations.

In addition, the expression for the mixture viscosity was derived for a dilute

suspension of liquid spherical droplets by Taylor [10]. Later, Roscoe derived

an expression for the mixture viscosity for solid spheres at higher volume

fraction [11]. The two expressions are combined in the mixture viscosity in

the Ishii-Zuber drag law. However, Taylor assumes spherical droplets, while in

the present experiments considerable deformations are seen. Furthermore, while

some deviations from the spherical shape are permitted according to Roscoe, his

derivations of the mixture viscosity assumes a very wide particle size distribution.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

206 Advances in Fluid Mechanics IX

from the assumptions in the model derivation also contribute to the observed

discrepancies in the results.

The kinetics of emulsion breakup were investigated with the same system that

was used for the compression experiments. Oil-in-water emulsions stabilized by

10 mM SDS are highly stable due to the colloidal forces that prevent droplets

from coalescing [12]. The addition of inorganic electrolytes is known to promote

coalescence in emulsions stabilized by ionic surfactants such as SDS, due to a

decrease of the electric double layer repulsion between the droplet interfaces [12].

One drop of emulsion with a volume of 350 nl and φ = 0.72 was added to the

sample chamber. The sample was then briefly centrifuged to push all liquid to the

bottom of the chamber. Then, a droplet of a 5 wt% NaCl aqueous solution (350 nl)

was added to the sample chamber. A snapshot of the sample chamber after addition

of the two solutions is displayed in Figure 4 (photo at the left). The concentration

of SDS after addition of the NaCl solution changed to 5 mM and the concentration

of NaCl to 2.5 wt%. The interfacial tension of Sil180 and an aqueous phase with

this composition was measured to be 5.4 mN m−1 . Emulsions prepared in this

way were centrifuged for 10 minutes at different accelerations in the range 64

< a/g < 2260. Figure 4 displays snapshots of an emulsion centrifuged at 864g at

different times. The droplets are compressed at the top of the emulsion. A layer

of pure oil, which grows with time, is visible after 1 minute of centrifugation.

As continuous phase is flowing from top to bottom in the emulsion due to the

enhanced gravitational acceleration, the film of continuous phase between droplets

is always thinnest in the top layer of the emulsion. A thinner film is more likely to

rupture, hence coalescence takes place mostly in the top layer of the emulsion.

In addition to coalescence between oil droplets and the pure oil homophase,

coalescence between droplets in the emulsion layer was also observed. This is best

visible in the snapshot in Figure 4 taken after 10 minutes. The extent of drop-drop

coalescence was not evaluated in this work, as the presence of a droplet bilayer

causes problems for automated image analysis.

The volume fraction of coalesced oil xc = Vc /Vt was evaluated. Vc and Vt are

the volumes of coalesced oil and total volumes of oil in the system, respectively.

Figure 5A displays three curves xc (t) as examples. Curve 1 was recorded at 2260g

acceleration, curves 2 and 3 were recorded at 1282g acceleration. For curve 1, xc (t)

increases approximately linearly until xc = 1 is reached, which indicates that all

oil from the emulsion has coalesced. From a linear fit to the data points for which

xc < 1 we calculate a mean coalescence rate rc = dxc /dt. For curve 2, xc < 1 for

the duration of the experiment. The difference in slopes between curves 1 and 2

indicates that the coalescence rate is larger for 2260g acceleration as compared

to 1282g. It was often observed that coalescence took place before centrifugation

was started, as a few seconds passed between addition of the NaCl solution and the

start of centrifugation. Curve 3 in Figure 5A displays an example where significant

coalescence occurred before centrifugation and thus the intercept of the linear fit

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 207

solution during centrifugation at 864g centrifugal acceleration. The

picture at the left displays the sample chamber after additions of 350 nl

emulsion and 350 nl NaCl solution. The other photos display the sample

chamber at different times of centrifugation. The sketch at the right

schematically displays the distribution of phases during centrifugation.

centrifugation time for three individual experiments as examples. The

dotted lines are linear fits to the data points. The curves are discussed in

detail in section 3.2.

xc (t) at t = 0 would deviate strongly from zero. The first data point was taken after

one minute centrifugation, xc (t) increased approximately linear from this point on.

The slopes of curves 2 and 3 differ significantly from each other, even though

experiments were performed at the same acceleration. The scatter of the data points

was generally large, which is visualized in Figure 6A displaying rc as a function

of the relative radial acceleration a/g obtained from individual measurements.

The reason for this large scatter is currently not known. For each value a/g,

5-7 individual measurements were performed and the average value rc was

calculated. Even though the error bars in Figure 6A are large, rc increases

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

208 Advances in Fluid Mechanics IX

Figure 6: Mean relative coalescence rate rc (panel A) and mean coalescence time

τ (panel B) between a droplet and the pure oil phase as a function of

the relative acceleration a/g. The open circles in Panel A are results of

individual measurements, the filled circles are average data points from

5-7 measurements. Error bars are 95% confidence intervals.

universal property, as it depends on the area of the pure oil/emulsion interface

Ai . A more general parameter is the characteristic coalescence time τ , which is the

time it takes for the thin film of continuous phase between the droplet and the pure

oil phase to drain and rupture. We have shown previously that τ can be calculated

as τ = Ai φt /4V0qt rc [5]. Ai is given by Ai = hw, where h and w are the channel

height and width, respectively. φt is the dispersed phase volume fraction in the top

layer of the emulsion, for which we will assume φt = 1. V0 is the total volume of oil

in the system, qt is the ratio of cross-sectional area Ad and volume Vd of a droplet

in the top layer of the emulsion, qt = Ad /Vd .

From the images we see that the droplets in the top layer are deformed. If we

assume the droplets to be in the shape of a rhombic dodecahedron, which is a

common approximation for deformed droplets in a dense emulsion layer, then we

obtain qt ≈ 2.29/d, where d is the diameter of an undeformed droplet of the same

volume. The coalescence time τ is displayed in Figure 6B as a function of the

relative radial acceleration a/g; τ decreases with increasing a/g.

In the simplest model for film drainage, a circular film forms at the droplet

interface which is parallel with the pure oil interface. The coalescence time τ is

then given by: τ = 12πηc r4f /Fh2c [13]. ηc is the continuous phase viscosity, r f the

film radius, F the compressive force and hc the critical film thickness, at which

coalescence will occur. The equation predicts a decrease of τ with increasing

compressive force, which was observed in our experiments.

Several factors will cause quantitative deviation of the experimental data from

this simple model, however. Firstly, the film radius itself will depend on the

compressive force, an estimate is given by r2f = Frd /2πσ , where rd is the droplet

radius and σ the interfacial tension [12]. Due to the confinement of droplets in a

dense emulsions, r f cannot increase arbitrarily with F, but will take on a maximal

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 209

value at a certain force. Secondly, the critical film thickness is mostly only known

as an order-of-magnitude estimate, and is thus a considerable source of error, as

τ ∝ h−2 . Thirdly, as the coalescing droplet is surrounded by a dense emulsion, the

flow of continuous phase may be significantly impeded due to the high hydraulic

resistance of the network of channels of continuous phase that is formed between

the emulsion droplets, as compared to film drainage between an isolated drop and

the pure oil interface. Fourthly, the experiments were performed with a micellar

solution as the continuous phase. The presence of micelles may further modulate

the interaction force between the droplets at distances < 100 nm [12], which is

difficult to quantify. Hence, a quantitative description of the coalescence dynamics

cannot be done in a reliable manner, and we restrict ourselves to the qualitative

discussion given above.

4 Conclusions

under enhanced gravity using in-situ microscopy. The novelty of the experimental

approach is that it allows direct optical monitoring of emulsion droplets during

centrifugation at high g-forces.

The agreement between the emulsion compression experiments and numerical

simulations was not good for two of the three employed radial accelerations.

Possible reasons for this are the loss of interfacial physics due to the averaging

procedure used for the two-fluid model and also the disagreement of the

experiments with some of the modeling assumptions. We would like to point out

that the results demonstrate that the Ishii-Zuber drag law, which is very popular,

should be employed critically in CFD calculations simulating multiphase systems

with high dispersed phase fractions and/or under conditions where enhanced

gravity would cause significant droplet deformation.

In the coalescence experiments it was found that the coalescence rate increased

approximately linearly with a/g. This behavior was explained qualitatively with a

simple film drainage model.

Improvements of the current experimental setup are possible for future work.

In order to study coalescence between droplets in the emulsion, a monolayer of

droplets must be produced to allow for a better visibility of individual droplets.

As we want to work with droplet diameters < 100 µm because of their relevance

for industrial applications, the height of the sample chamber needs to be ≤ 100

µm. When using PDMS as chamber material, we observed that channels of height

≤ 100 µm and width = 1 mm collapsed due to downward bending of the lid.

To avoid this, the sample chamber can be made from glass. The procedure of

destabilizing the emulsion by addition of salt may cause coalescence before the

start of centrifugation. The destabilization can be triggered in a more reproducible

way by choosing a photo-destructible surfactant, that will lose its emulsifying

capability upon irradiation of the sample with light [14]. This is expected to reduce

the statistical uncertainty of the measured coalescence rate.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

210 Advances in Fluid Mechanics IX

Acknowledgements

This work is carried out as part of a project of the Institute for Sustainable

Process Technology: development of an Ω2 R separator focusing on oil/water

separation (project number OG 00-04). We thank Laurens van Campen from Delft

University of Technology, Johanna Bos from Frames and Paul Verbeek from FMC

Technologies for helpful discussions.

References

[1] Ishii, M. & Zuber, N., Drag coefficient and relative velocity in bubbly, droplet

or particulate flows. AIChE Journal, 25, pp. 843–855, 1979.

[2] Schiller, L. & Naumann, A., Uber die grundlegenden Berechnungen bei der

Schwerkraftaufbereitung. Zeitschrift des Vereines Deutscher Ingenieure, 77,

p. 318, 1933.

[3] Rusche, H., Computational fluid dynamics of dispersed two-phase flows at

high phase fractions. Ph.D. thesis, Imperial College of Science, Technology

& Medicine, Department of Mechanical Engineering, London, UK, 2002.

[4] Jakobsen, H., Chemical Reactor Modeling: Multiphase Reactive Flows.

Springer, 1st edition, 2008.

[5] Krebs, T., Schroen, C. & Boom, R., Separation kinetics of an oil-in-water

emulsion under enhanced gravity. Chemical Engineering Science, 72, pp.

118–125, 2012.

[6] Qin, D., Xia, Y. & Whitesides, G., Soft lithography for micro- and nanoscale

patterning. Nature Protocols, 5, pp. 491–502, 2010.

[7] Krebs, T., Schroen, C. & Boom, R., A microfluidic method to study

demulsification kinetics. Lab on a Chip, 2012. doi:10.1039/C2LC20930F.

[8] Drew, D., Mathematical modeling of two-phase flow. Annual Review of Fluid

Mechanics, 15, pp. 261–291, 1983.

[9] Griggs, A., Zinchenko, A. & Davis, R., Gravity-driven motion of a

deformable drop or bubble near an inclined plane at low Reynolds number.

International Journal of Multiphase Flow, 34, pp. 408–418, 2008.

[10] Taylor, G., The viscosity of a fluid containing small drops of another fluid.

Proceedings of the Royal Society of London, Series A, 138, pp. 41–48, 1932.

[11] Roscoe, R., The viscosity of suspensions of rigid spheres. British Journal of

Applied Physics, 3, pp. 267–279, 1952.

[12] Birdi, K., Handbook of Surface and Colloid Chemistry. CRC Press, 2nd

edition, 2003.

[13] Chesters, A., The modelling of coalescence processes in fluid-liquid

dispersions: A review of current understanding. Trans IChemE, 69A, pp.

259–270, 1991.

[14] Eastoe, J., Photo-destructible surfactants in microemulsions. Progress in

Colloid and Polymer Science, 133, pp. 106–110, 2006.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 211

two-phase flow in a nozzle

S. Martel1, Y. Mercadier1 & M. Dostie2

1

Department of Mechanical Engineering, Université de Sherbrooke,

Canada

2

Laboratoire des technologies de l’énergie, Hydro-Québec, Shawinigan,

Canada

Abstract

This paper presents a steady state two-phase flow model including new choking

criteria for one-dimensional conservative systems. As a first step, this model is

used to study the flow in the motive nozzle of an ejector. Mechanical

disequilibria and momentum exchange between phases are taken into account

and the numerical scheme uses the SIMPLE algorithm. Numerical results are

compared to experimental and previous numerical results from the literature.

Keywords: critical two-phase flow, critical location, steady state model.

1 Introduction

The steady state flow of compressible fluid through convergent-divergent

nozzles covers various important flow phenomena like the occurrence of critical

flow conditions, transition from subsonic to supersonic flow or the occurrence of

flow discontinuities. One phenomena still of interest is the choking condition in a

critical flow such as in supersonic ejectors. Supersonic ejectors are widely used

in a range of applications such as aerospace, propulsion, refrigeration and many

thermal systems. In this paper, a one-dimensional compressible steady state two-

phase flow model is presented with new choking criterions that are directly

related to optimal flux conditions developed by [1]. As a first step, this model is

used to study the flow in the motive nozzle of an ejector.

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120191

www.engbookspdf.com

212 Advances in Fluid Mechanics IX

2 Numerical model

This is a one-dimensional two-fluid single pressure model using one temperature

and mechanical disequilibria between phases in which no phase changes

occurred. The procedure used for the calculation of the flow field is the standard

SIMPLE algorithm developed for steady state flow conditions [2]. A constant

liquid density and perfect gas law for compressible component are assumed. In

addition, droplets in gas flow for momentum exchange between phases are

assumed to be uniform and with constant diameter. In order to take account of

the mechanical disequilibria, the slip ratio between phases must be obtained

using drag source term.

With these assumptions, the following four equations model is obtained:

x k

f Mk as11u1 0 (1)

f Mk sk as11u12 d as p p as (2)

x k dx x

C P

2

1 1 1 Mk k

A u f s 2 u1

f Mk fTk p1 0 (3)

x k 2 k R11

x

x

A 2 2u 22 as 2 p 2, J 2c, J (4)

where 2,J is the drag force between phases and 2c,J is the force cause by the

variable section:

V2 A2D C D

2, J 1 u 2 u1 u 2 u1 (5)

V p x 2

2c, J p a s 2 (6)

x

In these equations, sk is the slip ratio, fMk the mass flow ratio, and fTk enthalpy

ratio of phase k. These ratios can be respectively defined by:

uk

sk (7)

u1

ak k u k

f Mk (8)

a1 1u1

Cp k Tk

f Tk (9)

Cp1T1

In the previous equations, as is the sectional area, k the volumetric fraction of

each phase k, k the density of each phase, uk the velocity of each phase, p the

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 213

flow pressure, Cp1 the heat capacity of compressible phase, and R1 the specific

gas constant of compressible phase. For compressible phase, k is 1 and for

incompressible phase, k equal 2. Volumetric fraction and geometrical section are

linked by the following equations representing a global constraint that all phases

must occupy the total nozzle section:

ak

k

as (10)

k 1

k

3 Validation

At the inlet of the nozzle, homogeneous equilibrium flow is assumed from

reservoir. For this reason, each phase have the same temperature and pressure.

The inlet volume fraction is imposed. For the outlet boundary condition a

constant exit pressure is applied. The initial conditions in the nozzle are identical

with the upstream reservoir conditions which imply that the transient calculation

starts with a strong discontinuity at the nozzle exit. The model has been validated

with experimental data from Elliot and Weinberg [3] and with Carofano and

McManus [4]. Finally, results have been compared to numerical results from

Städtke [5].

show one pressure profile and a table resuming experimental results obtained for

different mass flow ratio such as mass flow rate, exit velocity, and thrust

measurement. Inlet conditions for all tests are p0= 10.3421 (MPa), T0=293K and

an exit pressure of 0.1013 MPa. The Figure 1 shows that numerical pressure

1.2

P exp

Pressure (MPa)

1.0 P SIMPLE

Throat

0.8

0.6

0.4

0.2

0.0

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4

Figure 1: Pressure profile in the nozzle.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

214 Advances in Fluid Mechanics IX

profile is in good agreement with experimental results. For this test, mass flow

ratio is 39.1 and experimental mass flow rate is 67.33 kg/s. The numerical mass

flow rate obtained is 67.64 which give an error of 0.45%.

Table 1 shows comparison between numerical and experimental results for

mass flow ratio range of 15.3 – 64.9. For all tests, results are in good agreement

for mass flow rate since error varies from 0.11 to 2.77%. However, comparison

of experimental and numerical thrust results give an overestimation in all cases

between 4.96 to 8.28%. From Elliot and Weinberg numerical results, the

difference between wall frictionless flow velocity and the velocity with friction

is about 5.6%. From this observation, the wall frictionless flow assumption

resulting in a higher mean velocity used in the calculation of the thrust is the

cause of this overestimation.

M Thrust M Error Error Thrust Error

flow velocity velocity

(kg/s) (N) (kg/s) (%) (%) (N) (%)

ratio (m/s) (m/s)

15.3 44.5 143 6334 45.8 2.77 147 3.35 6748 6.54

17.2 47.7 131 6236 48.1 0.83 140 7.07 6753 8.28

21.1 51.8 125 6441 52.5 1.27 129 3.34 6760 4.96

22.3 53.6 118 6308 53.7 0.14 126 6.73 6762 7.21

28.3 57.7 109 6308 59.3 2.76 114 4.34 6773 7.38

30.1 60.9 104 6334 60.8 0.11 111 6.83 6776 6.97

39.1 67.7 94 6334 67.3 0.59 100 7.30 6760 6.72

51.6 74.5 85 6334 75.4 1.13 90 6.27 6813 7.55

64.9 81.4 78 6334 82.2 1.00 83 6.73 6844 8.04

The nozzle used by Carofano and McManus [4] to obtain experimental two-

phase flow results is used to validate numerical results. They obtain a pressure

profile for p0 = 0.3440 MPa, T0 = 289 K and an exit pressure of 0.1455 MPa. The

mass flow ratio used in this case is 0.1013. The experimental mass flow rate is

0.4141 kg/s and the numerical mass flow rate obtained is 0.4103 which implies a

difference under 1%. The pressure profile is in good agreement with

experimental.

numerical results. For this case, fixed upstream reservoir pressure and

temperature of p0 = 1 MPa, T0 = 400 K, u1 = u2 is used with a mass flow rate

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 215

ratio (FMk) of unity and an exit pressure of 0.6 MPa. Numerical results obtained

with this scheme are in good agreement with those of [5] as shown on Figure 1.2.

Total mass flow rate obtained by [5] is 5.68 kg/s compared to 5.70 kg/s obtained

with this model which implied a difference of 0.35%. It is interesting to see that

the numerical scheme is able to capture choc wave in the nozzle. The differences

between velocity profiles can be explained by the fact that no information about

the droplet assumptions are given by Städtke. Droplet diameter and drag

correlation have a direct effect on the slip ratio and the velocity profiles. In

addition, Städtke are using a two temperatures model giving some differences in

the phase velocity results.

0.4

Pressure (MPa)

0.3

P exp

P SIMPLE

0.2 Throat

0.1

0.0

0.00 0.03 0.06 0.09 0.12

Axis position (m)

1.2

Pi (Pa)

Pressure (MPa)

1.0 Throat

P Stä dtke

0.8

0.6

0.4

0.2

0.0

0.0 0.2 0.4 0.6 0.8 1.0

Axis position (m)

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

216 Advances in Fluid Mechanics IX

600

U1

500 U2

Throat

U1 - St ä dtke U1

400 U2 - St ä dtke

300

U2

200

100

0

0.0 0.2 0.4 0.6 0.8 1.0

Axis position (m)

4 Conclusions

Multiphase systems present significant scientific challenges and the choking

phenomenon has to be taken into account properly to achieve an accurate model

of two phase flow. More results are obtained concerning super-critical behavior

in comparison with mono-phase compressible flow. This numerical scheme give

good results and the numerical study have given a lot of information about

critical two-phase flow behavior. It also shows that steady state models are

achievable for critical two-phase flow. Such an approach is in some cases

simpler that using transient models which involve much more complex

propagation phenomenon. This model allows a good range of inlet void fraction

from droplet flow (void fraction close to 1) to bubble flow (void fraction close to

0.4). Numerical results also validate the fact that critical location is not necessary

at the throat section. For variable slip ratio, this is not necessary the case since

mass, momentum and energy exchanges between phases give a different

behavior and critical location appear after the throat. Numerical study and

theoretical consideration are into progress to predict critical position and effect

of mass flow ratio and thermical disequilibria on flow behavior.

Acknowledgements

The work described here was supported by NSERC, FQRNT and NSERC Chair

in Industrial Energy Efficiency (chairholder: N. Galanis) at Université de

Sherbrooke with the support of Hydro-Quebec (Energy Technology Laboratory,

LTE), Rio Tinto Alcan and the CANMET Energy Technology Center (CETC-

Varennes, Natural Resources Canada).

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 217

References

[1] Dostie M., Mercadier Y., Critical fluxes in conservative systems: application

to multiphase flow. Submitted to Journal of Fluid Mechanics, 2011.

[2] Patankar, S.V. (1980). Numerical heat transfer and fluid flow, Hemisphere

Publishing Corporation, New-York, 197 p.

[3] Elliot, D. G. and Weinberg, E. Acceleration of Liquids in Two-Phase

Nozzles, JPL Tech. Rep. 32-987, 1968.

[4] Carofano, G. C. and McManus, H. N. (1969). An analytical and

experimental study of the flow of air-water and steam-water mixtures in a

converging-diverging nozzle, Progress in Heat and Mass Transfer, Vol. 2,

Pergamon Press, Oxford.

[5] Städtke, H. Gasdynamic Aspects of Two-Phase Flow. WILEY-VCH,

Weinheim, 273 p, 2006.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

This page intentionally left blank

www.engbookspdf.com

Advances in Fluid Mechanics IX 219

boiling flow in a vertical pipe with phase

change using a multi-fluid modelling approach

R. Kopun1 & L. Škerget2

1

AVL-AST d.o.o., Maribor, Slovenia

2

Faculty of Mechanical Engineering, University of Maribor, Slovenia

Abstract

Multi-phase boiling flow inside a vertical pipe is simulated as part of a

preliminary study of the heat transfer characteristics of coolant flow inside water

cooling jackets in IC engines. Based on increasing demands for higher efficiency

inside the engine cooling block, heat transfer plays an important role in a

conceptual and thermal analysis used to provide efficient cooling. The

simulations of thermal and flow phase change characteristics inside a vertical

pipe boiling system were made to prevent component failure and to create a

uniform temperature distribution inside the water cooling jacket. The developed

boiling mass transfer model, such as the BDL (Boiling Departure Lift-off)

model, has empirically correlated heat transfer coefficients and is implemented

within the commercial computational fluid dynamics code AVL FIRE®.

Governing equations are based on the Eulerian multi-fluid approach which treats

each phase as interpenetrating continua coexisting in the flow domain, with

inter-phase transfer terms accounting for phase interactions. Turbulence is

modelled by using an advanced - - model. In this paper, we focus on

comparison and suitability of the mentioned boiling model with two different

boundary condition approaches to determine the HTC (heat transfer coefficient)

for multi-phase boiling flow inside the vertical pipe. Temperature measurements

along the height of the pipe were performed at three different positions.

Comparison with the available experimental data for different boundary

conditions is presented. Simulation results exhibit good agreement with the

experimental data.

Keywords: multiphase flow, boiling, vertical pipe, HTC, CFD, IC engines.

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120201

www.engbookspdf.com

220 Advances in Fluid Mechanics IX

1 Introduction

Detailed knowledge of flow and heat transfer analysis is essential in order to

achieve a controlled boiling over a wide range of working conditions. Heat

transfer applications such as those introduced in automotive industries often play

an important role in conceptual and thermal analysis of cooling system

(Dong et al. [1]). Vapour bubbles usually start growing in superheated zones

near the walls and are normally able to condense harmlessly back into the cooler

liquid, but under certain conditions (flow geometry, fluid conditions, est.)

Transition from nucleate to film boiling regime occurs. That kind of behaviour

dramatically reduces heat transfer rate which leads to large rise (upturn) of

component temperature and contributes to the failure of the device. Hence,

efficient cooling is essential to prevent component failure and provide even

temperature distribution. This leads to reduction of thermal stresses and higher

durability and has consequently influence on power requirements

(Campbell et al. [2]).

Modelling of flow boiling heat transfer usually relies on the empirical

approaches. It is based on the superposition of convective and boiling

components, as first suggested by Rohsenow, i.e. represented by simple addition

of the nucleate and convective coefficient. This approach was afterwards

extended by Chen (Steiner and Taborek [3]). In recent years many new

approaches and correlations have been proposed. One of them is the newly

developed model for mass transfer function based on the assumption that the

mass transfer parameters are proportional to the heat transfer coefficient from the

fluid system as proposed by Srinivasan and Wang [4]. The model has already

been successfully implemented and applied to the quenching boiling heat

transfer process showing significant improvement as reported by Wang et al. [5].

Heat transfer of coolant flows inside water cooling jacket often involves a

phase change which can be simulated, as a preliminary study, with the horizontal

and the vertical pipe for easier explanation of flow characteristics. Recently,

Srinivasan [6] has developed the new mass transfer model to simulate thermal

and phase change characteristics for binary mixture during flow boiling process

inside a horizontal channel. Good agreements of numerical predictions with

experiments are presented and the applicability of developed model shows that it

can be easily extended to automotive applications.

In this work we simulate the multi-phase boiling flow inside a vertical pipe

with commercial CFD code AVL FIRE. A brief overview of the governing

equations of the newly developed mass transfer model is described first.

Boundary conditions, numerical setups and simulation results are presented and

compared with the available experimental values. Results are discussed and

summarized in the concluding.

2 Numerical background

Eulerian multi-fluid methods consider each phase as interpenetrating continua

coexisting in the flow domain, with inter-phase transfer terms accounting for

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 221

phase interactions where conservation laws apply. From the theoretical work of

Lahey and Drew [7] the averaged continuity and momentum equations are

presented as follows:

2.1 Continuity

∙ Γ 1, … , (1)

,

where , and v stand for volume fraction, density and velocity. Phase change

rate (in this particular case, boiling) is Γ and the subscript is a phase indicator

( or ).

2.2 Momentum

v Γ (2)

while , and v are respectively the pressure, stress and interfacial velocity.

Interfacial momentum transfer term with the drag is being the most important

force and due to boiling initiated along the solid and liquid interface, mass

interfacial exchange occurs. Interfacial momentum exchange term is given as:

1

| | (3)

8

is the relative velocity. Subscripts and denote the continuous and

dispersed phases of the given flow. Drag of coefficient is defined as

24

1 0.15 . 1000 v

(4)

0.438 1000

Presented model does not account for the turbulence dispersion force on the

momentum interface. In the framework of the two-fluid model, an individual

energy equation is be solved for each phase, where it is assumed that two phases

are in thermal equilibrium, because the heat transfer rate between vapour and

liquid phase is relatively rapid. Further details concerning modelling interfacial

mass exchange can be obtained from AVL FIRE Multi-fluid model, solver

theory guide [8].

2.3 Energy

inside the code to model the effects of turbulence within the multiphase system.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

222 Advances in Fluid Mechanics IX

(5)

Γ

, ,∉ , ,∉

with

(6)

between phases and is denoted as and , and represent volume

fraction, dynamics viscosity and conductivity, respectively heat flux is given

by

(7)

,

where , is the mixture specific heat. Eqn. for turbulent kinetic energy and

dissipation rate are given by

(8)

Γ

, ,

(9)

Γ v

, ,

equations can be obtained from AVL FIRE Multi-fluid model, solver theory

guide [8].

Based on the assumption that the heat transfer rate is proportional with the phase

change rate, since the mass transfer predominantly controls heat transfer, the

phase change rate due to boiling process can be written as

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 223

∙ ∙ ∙ ∙∆ (10)

Γ

coefficient, the boiling heat transfer coefficient and the latent heat of

vaporization, respectively. Wall superheat temperature ∆ , the interfacial area

density and the closure coefficient used to correct the interfacial area

density in eqn. (10) are defined by

6

∆ 1 (11)

with being the dispersed dry phase packing limit and being the

minimum volume fraction. The boiling correction coefficient is imposed to

correlate nature of boiling process such as film, partial nucleate, transition

boiling model etc. The heat transfer coefficient in the eqn. (10) is used for

computing the mass transfer exchange rates and assumes the computed value of

boiling heat transfer coefficient for different boiling regimes. Heat transfer

coefficient for binary mixture is evaluated with the Chen correlation [10] as

∙ (12)

. . .

∙ ∙ .

0.00122 . . . .

∙ ∙ ∙ (13)

.

∙

refer to liquid thermal conductivity, liquid specific heat at constant

pressure, surface tension, liquid dynamics viscosity, latent heat of vaporization,

wall temperature, bubble point temperature, saturation pressure at and

saturation pressure at .

Recent modification of the Chen model were made by Steiner et al. [11] with

combination of Zeng et al. [12] and the model was fully implemented into FIRE

as the BDL model. The main difference between the two models is in

determination of suppression factor where the Chen model cannot take local

fluid state into account as a function of a reference Reynolds number of the

global geometry. The BDL model is a recent improvement in which the

suppression factor is computed from local velocity and length scales. The boiling

suppression factor used in eqn. (12) is decomposed in two parts as

∙ (14)

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

224 Advances in Fluid Mechanics IX

and growth and the second suppression factor correlates correction to the

defined heat transfer coefficient modelled in eqn. (14) as

∙

(15)

1 ∙

calculated as a function of local velocity and differences in saturation to local

cell temperature, where and in eqn. (19) stand for local coefficient and

Nusselt number. Detailed information about parameters can be found in AVL

FIRE Multi-fluid model, solver theory guide [8].

The present study consists of the experimental and numerical investigation on

the vertical pipe boiling case. The experimental setup is schematically

demonstrated in Fig. 1.

direction.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 225

The total length of the pipe is 1.9 m with the outer diameter of 0.03 m.

Heated section, length of 1.5 m with the inner diameter of 0.012 m, has a

measured constant heat flux of 44139 W/m2. At the inlet and outlet there are

small parts of the pipe with length of 0.2 m which are blind, i.e. not exposed

to heat flux.

There are three thermo-sensors known as IEC 584, tip-K with working

temperature ranged from 200 to 800 with measurement deviations of

1.5 or 0.004 ∙ | |. Thermo-sensors , and are placed on three

different positions along the pipe, where length in z direction to is 0.415 m,

0.885 m to sensor and 1.35 m to sensor . Inlet water is preheated to the

temperature of 80 and it enters the inlet domain with different velocities. The

volumetric flow range varied from 1.04 / to 2.18 / ,

corresponding to velocities from v 0.04148 / to v 0.08695 / .

Mixture of two-phases (water and steam) exits the vertical pipe at the top with

static pressure of 1 bar. Data measurements were perform at the surrounding

temperature of 21 .

Basic elements of the numerical model configuration, shown in Fig. 2,

include flow inlet and outlet section, heated wall, non-heated wall and wall.

Heated wall area (in Fig. 1 marked with parameter ) selection is available to

present constant heat flux 44139 / as in the original experimental set

up, where the wall boundary condition consists of a natural convection to

surrounding temperature 21 with the heat transfer coefficient

10 / . The non-heated wall is modelled with 0 / adiabatic

boundary condition. The entire domain is meshed by using a hexahedral type

structure to a count of 160,000 cells.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

226 Advances in Fluid Mechanics IX

At the pipe inlet the fluid velocity is set only for z-direction (in our case in the

liquid flow direction) with different velocities v and v with preheated water to

80 as it is described in the experimental set up. The minimum volume fraction

admissible for each phase in the entire domain is set to 1 ∙ 10 and phase one

stands for water and phase two is vapour. Transient simulations with different

time steps are conducted to compare with the experiment which lasted for 100

sec.

The numerical simulations were performed with the commercial CFD code

AVL FIRE® v2010.1 in which the Finite Volume approach was used to solve

governing equations and where numerical solution procedure was based on

SIMPLE algorithm extended for a multiphase flow case. The normalized residual

limit for mass, momentum, and volume fraction were set to 2 ∙ 10 , while the

turbulence was allowed to drop until it reaches 1 ∙ 10 , where the energy

relaxation was extended to the value of 1 ∙ 10 . Turbulence was modelled with

an advanced - - model, where homogeneous turbulence interface exchange

between different phases was used.

a) b)

0.04148 / with experiment for different times, a) t 1

and b) t 100 with zoom area of the thermocouples ,

and .

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 227

4 Simulation results

The 3D results presented in the following section focus around the monitoring

points , and for easier visualization purpose and to observe the flow

variable and physic effects on details between numerical and experimental

results.

The effect of constant heat flux on the growth and propagation of volume

fraction, for a given inlet velocity condition v 0.04148 / , can be

identified from Fig. 3. It can be seen that after t 1 (Fig. 3a) no vapour

bubble is generated because the heat flux produces a temperature that is lower

than the saturation temperature. By a close look at the experimental results in

Fig. 3a, it is observed that there is also no vapour bubbles generated, whereas at

the time t 100 (Fig. 3b) vapour bubbles are initiated in the heated wall

area section and develops to the upstream of the tube.

Good agreement between experimental and numerical results can be observed

in Figs. 3a and b. From the Fig. 3b at t 100 of heating time, it can be

seen that heavy boiling occurs in area by zooming thermocouples , and

areas. It is found that the temperature has almost reached the saturation

temperature in the area of as some bubbles can be observed in the heated

area, whereas in the area no vapour bubbles are observed which implies that

the temperature at this height is lower than the saturation temperature. Similar

results were generated with other inlet velocity conditions (v 0.08695 / ).

The results are not presented here.

Detailed comparison of experimentally measured temperatures and the

corresponding simulated ones for two different inlet velocity cases (v

0.04148 / and v 0.08695 / ) on 3 different positions along the height

of the pipe are displayed in Figs 4 and 5. Fluctuations of the experimental values

appearing in Figs. 4 and 5 could be partly caused by noise during the

measurement as expected.

At the lower inlet velocity (v 0.04148 / ), Fig. 4 shows the deviation

in temperature predicted by the model against the measured values, noted for all

3 positions along the height of the pipe in transition area, between the beginning

and before calculation reaches a steady state. In general, the model captures the

trend of the experimental temperature histories. It can be seen that when

calculation reaches a steady state (t 100 ), for the position of thermocouple

, it has just the absolute deviation of ∆ 0.8 from the experiment, and

deviations of ∆ 0.2 and ∆ 0.5 for the thermocouples and ,

respectively. The maximum deviation for thermocouple is ∆ 5.7 at

around t 5 sec, whereas the deviation for the thermocouple is ∆ 5.9 .

Similar results are obtained also with higher velocity (v 0.08695 m/s, see

Fig. 5), where the difference between the measured and simulated values are

presented at the same positions along the height of the pipe. Good agreement is

obtained in this case. The maximum deviation reaches ∆ 3.5 at t=100 s

(relative deviation of 1%) in the area of thermocouple . When the calculation

reaches a steady state in the area of thermocouple , the absolute deviation is

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

228 Advances in Fluid Mechanics IX

predicted by the numerical model (red dashed line) against the

experimentally measured values (black solid line) with inlet liquid

velocity v 0.04148 / .

deviation appears in developing stage at around t 5 and has a value of

∆ 4.9 for the area of thermocouples .

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 229

predicted by the numerical model (red dashed line) against the

experimentally measured values (black solid line) with inlet liquid

velocity v 0.08695 / .

5 Conclusion

within the commercial CFD code AVL FIRE® is capable of predicting boiling

under different conditions in a vertical pipe. It can be concluded that all

numerical results show a good agreement with the available experimental data.

While better trends and lower deviations were observed for lower inlet velocities,

higher inlet velocities can have a maximum relative deviation of 1.4% during the

whole boiling process. In the current contest all calculations were performed

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

230 Advances in Fluid Mechanics IX

with the advanced - - turbulent model, but the forthcoming work will

perform present simulation with the variety of turbulence models and different

inlet velocities to identify flow pattern predictions. Comparison of the vapour

size and volume fraction with measurements will be performed. The

computational method and workflow discussed in this article are capable of

reporting the temperature values in combination with phase characteristics in the

flow domain and can be used as a preliminary study of heat transfer

characteristics of coolant flow inside water cooling jackets in internal

combustion engines.

References

[1] Dong F., Fan Q., Cai Y., Jiang S., Guo C., Norihiko W., Lee W.T.,

Numerical simulation of boiling heat transfer in water jacket of DI engine,

SAE International, 2010.

[2] Campbell N.A.F., Charlton S.J., Wong L., Designing towards nucleate

boiling in combustion engines, IMechE C496/092, 1995, pp. 587-595.

[3] Steiner D., Taborek J., Flow boiling heat transfer in vertical tubes

correlated by an asymptotic model, Heat Transfer Engineering, Vol. 13,

No.2, 1992, pp. 43-69.

[4] Srinivasan V., Wang D.M., Modelling and simulation of the heat and mass

transfer characteristics of binary mixtures for boiling flow applications, J.

ASTM international, Vol.8, No.2, 2011, pp. 1-22.

[5] Wang D.M., Alajbegovič A., Su X.M., Jan J., Numerical simulation of

water quenching process of an engine cylinder head, Proceedings of ASME

FEDSM 2003, 4th ASME JSME Joint Fluids Engineering Conference,

Honolulu, Hawaii, USA, July 6-10, 2003.

[6] Srinivasan V., Numerical simulation of flow boiling of binary mixtures

using multi-fluid modelling approach, ASME international Mechanical

Engineering Congress and Exposition, Colorado, 2011, pp. 1-12.

[7] Lahey R.T.Jr., Drew D.A., An Analysis of Two-Phase Flow and Heat

Transfer Using a Multidimensional, Multi-Field, Two-Fluid Computational

Fluid Dynamics (CFD) Model, Japan/US Seminar on Two-Phase Flow

Dynamics, Santa Barbara, California, 2000.

[8] AVL LIST GmbH, FIRE CFD Solver, Eulerian Multi-fluid model. Solver

Theory Guide. 2010, Graz, Austria.

[9] Hanjalic K., Popovac M., Hadziabodic M., A robust near-wall elliptic-

relaxation eddy-viscosity turbulence model CFD, Int. J. Heat and Fluid

Flow, 25, 2004, pp. 1047-1051.

[10] Chen J.C., Correlations for boiling heat transfer to saturated liquids in

convective flow, I&EC Process Design and Development, Vol.5, No.3,

1966, pp. 322-329.

[11] Steiner H., Kobor A., Gebhard L., A wall heat transfer model for subcooled

boiling flow, Int. J. Heat and Mass Transfer, vol.48, 2005, pp. 4161-4173.

[12] Zeng L.Z., Klausner J.F., Bernhard D.M., Mei R. A unified model for the

prediction of bubble detachment diameters in boiling systems-II. Flow

boiling, Int. J. Heat Mass, Vol. 36, No.9, 1993, pp. 2271-2279.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 231

flow behaviour in fluidized bed reactors

R. K. Thapa & B. M. Halvorsen

Department of Process, Energy and Environmental Technology,

Telemark University College, Norway

Abstract

The aim of this study is to develop a robust CFD model for predicting fluid

dynamics in a gasification reactor. Experimental tests are performed. A

cylindrical bed with pressure sensors is used in the experimental study. A series

of simulations are performed using the commercial CFD tool ANSYS Fluent

12.1. A multi-fluid Eulerian model incorporating the kinetic theory of granular

flow is applied in the simulations. Fluidized bed reactors in biomass gasification

processes use steam as a fluidizing gas. High temperature makes it difficult to

study the flow behaviour under the operating conditions. A cold flow model is

constructed to study the fluid dynamics. Air at ambient conditions is used as the

fluidizing gas for the cold model. The density and viscosity variation between air

at ambient temperature and steam at high temperature results in different flow

behaviour. The CFD model is developed to also be able to predict the flow

behaviour of steam fluidized beds. Computational minimum fluidization

velocity, bed expansion ratio pressure drop and pressure standard deviation agree

well with experimental measurements. A computational model has been

developed and validated against experimental data. The validated CFD-model

can be useful in the study of flow behaviour of high temperature steam fluidized

gasification reactors.

Keywords: fluidized bed, CFD, multi-fluid Eulerian model.

1 Introduction

Gas-solid fluidized bed reactors are widely used in biomass gasification

technology. The fluidizing gas in the reactors is steam at high temperature. Study

of the flow behaviour inside the hot bed is difficult. Most of the design and

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120211

www.engbookspdf.com

232 Advances in Fluid Mechanics IX

Alternatively, down-scaled pilot plants and cold flow models are constructed. In

the cold flow model, air at ambient conditions is used as the fluidizing gas. Air at

ambient condition and high temperature steam has different density and

viscosity. The density and viscosity effect in the flow behaviour and fluidization

properties. The computational fluid dynamics (CFD) offers an approach to the

prediction of the flow behaviour. A validated model can be applied in the

simulation of flow behaviour in a hot steam fluidized reactor. This can reduce

the task of constructing pilot-scale and cold flow models.

Eulerian multi-fluid model is becoming more and more accepted in gas-solid

fluidized bed simulations [1]. The model incorporating kinetic theory of granular

flow considers both gas and solid as interpenetrating fluids. The objective is to

establish a comparatively validated model. This requires agreement between the

experimental and simulated results on parameters such as pressure drop, pressure

standard deviation, bed expansion, minimum fluidization velocity and bubble

behaviour.

A number of works on validation of the model have been published [2–6]. A

reasonably good agreement between experimental and computational results is

reported. The governing equations are mass and momentum balance. However,

different drag models and constitutive equations have been used in the models.

A set of the constitutive equation with drag model has been finalized by

Jayarathna, S.A. [7]. This option is applied in this work. The computational

analysis of minimum fluidization velocity and bubble behaviour using pressure

standard deviation and fluctuation of solid volume fraction has been introduced.

2 Experimental set up

Experiments are performed in a Plexiglas cylinder with 1.4 m height and 0.084

m diameter. The pressure sensors are located along the height of the cylinder and

connected to the lab-view program for data storage as shown in Figure 1.

digital flow controller, pressure sensors.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 233

The air flowing through a uniform air distributer is controlled by the lab-view

program in order to maintain a steady flow of gas. The air flow rate is controlled

by air flow meter and the data are saved in the lab-view program.

The particles in biomass gasification reactors are quartz sand of mean particle

size 500 µm and density 2500 kg/m3. Similar particles are selected in the

experimental work.

The physical properties of gas and particles used in the experiments are

presented in Table 1. A series of experiments are performed for a wide range of

superficial air velocity.

Particle density [kg/m3] 2500 Glass

Gas density [kg/m3] 1.225 Air

Gas viscosity 1.78x10-5 Air

Particle diameter [µm] 500 Mean

Initial bed height [m] 0.32

The pressure sensors are located at 0.03, 0.13, 0.23, 0.33 m above the air

distributor as shown in Figure 2.

3 Computational model

A multi-fluid Eulerian model incorporating kinetic theory of solid particles is

applied to simulate the transient behavior of the bed. The governing equations

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

234 Advances in Fluid Mechanics IX

are the conservation of mass and momentum. The kinetic theory of granular flow

considers the conservation of solid fluctuation energy [8]. Simulations are

performed with air as fluidizing gas. The particles and properties are consistent

to those used in the experiments. The simulation parameters used for the model

are presented in Table 2.

Particle density [kg/m3] 2500 Glass

Gas density [kg/m3] 1.225 Air

Gas viscosity [Ps.s] 1.78x10-5 Air

Particle diameter [µm] 500 Mean

Restitution coefficient 0.9

Initial solid packing 0.6

Maximum solid volume fraction [-] 0.63

Bed diameter [m] 0.084

Static bed height [m] 0.32

Time step 1x 10-3

Number of iterations per time step 40

The combination of model is validated against the experimental data for different

flow conditions [7].

Grannular Bulk Viscosity Symlal O’Brien

Frictional Viscosity Constant

Frictional Pressure Based-ktgf

Solid Pressure Ma-ahamadi

RadialDistribution Function Ma-ahmadi

Minimum fluidization velocity is regarded as the most important parameter in

the design of fluidized bed reactors [9]. Minimum fluidization velocity (umf) is

experimentally determined plotting average pressure drop across the bed height

as a function of superficial air velocity as shown in Figure 3. Experimental

measurement of minimum fluidization velocity for the glass particles is about

0.24 m/s. The pressure drop is 129 mbar and the bed expansion ratio is 1.03. The

pressure drop is proportional to the gas velocity below minimum fluidization

condition.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 235

Average pressure drop [mbar/m]

Superficial gas velocity [m/s]

velocity.

deviation at different level of the bed height. After fluidization conditions, the

pressure fluctuation across the bed increases significantly. Pressure standard

deviation can be used to determine the minimum fluidization velocity from

experimental and computational results. Before fluidization the pressure standard

deviation is about zero. It increases with the particle movement in the bed. The

experimental and simulated minimum fluidization velocities are about 0.24 and

0.26 m/s respectively. The experimental umf is similar to that in Figure 3. The

deviation between experimental and computational umf is 8%. The increase in

pressure standard deviation with increasing superficial air velocity indicates

increasing bubble frequency. The Figure indicates a good agreement of

experimental and simulated results concerning the minimum fluidization

velocity.

Experimental(h=0.13m) Experimental(h=0.23m)

Computational(h=0.13m) Computational(h=0.23m)

Pressure standard deviation [‐]

Superficial gas velocity [m/s]

deviation as a function of superficial air velocity. Bed height =

0.13m.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

236 Advances in Fluid Mechanics IX

Figure 5. The contours are at different superficial air velocity and at 8s of real

time simulation. Bubbles start to appear at the superficial air velocity 0.28 m/s.

The predicted minimum bubbling velocity is slightly higher than minimum

fluidization velocity. The particles used in the study are characterized as Geldart

B particles. For Geldart B particles, the minimum fluidization and minimum

bubbling velocities are almost the same [9]. Bubble frequency is increased with

increasing air velocity. Small bubbles are formed at the bottom of the bed and

the size is increased as they rise along the bed height.

velocities.

The minimum bubbling velocities are also studied from the plot of time

average solid volume fraction from simulation data at the bed height 0.13m and

0.23m. Figure 6 shows the solid volume fraction variation against time at the bed

height of 0.23 m/s.

u=0.24m/s u=0.26m/s

Solid volume fraction [‐]

Time[s]

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 237

At the air velocity of 0.24 m/s, the solid volume fraction is almost constant as

shown in Figure 6. Slight fluctuation of the solid volume fraction starts at the air

velocity of 0.26m/s indicating the inception of fluidization.

The solid volume fraction at the superficial air velocity 0.28 m/s shows

significant fluctuation indicating bubble formation. The bubble frequency is

increased at the air velocity 0.30m/s.

u=28 m/s u=0.30m/s

Solid volume fraction [‐]

Time [s]

bed height of 0.13m.The pressure drop show good agreement at the gas velocity

above the minimum fluidization. The deviation between the computational and

experimental results is 7% at minimum fluidization. Above the minimum

fluidization condition, the deviations are less than 7%. However, below

minimum fluidization the deviation is significant. Eulerian multiphase model

considers both the solid and gas as fluids. The model considers the bed fluidizing

even at the gas velocity lower than minimum fluidization.

Experimental Simulated

Pressure drop [mbar]

Superficial gas velocity [m/s]

function of air velocity.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

238 Advances in Fluid Mechanics IX

simulations. In both cases the bed expansion ratio is increased consistently with

superficial air velocity. Simulated bed expansion is slightly lower than

experimental. The average deviation of bed expansion between computational

and experimental results is about 5%. At the higher air velocities, the

experimental bed expansion is an average expansion. The bed height in the

experiments is fluctuating and not constant at higher air velocities.

Experimental Simulation

Bed expansion ratio [‐]

Superficial air velocity [m/s]

superficial air velocity.

5 Conclusions

A multifluid Eulerian model incorporating the kinetic theory of granular flow is

applied for computational prediction of gas-solid flow behaviour. The results are

compared with the experimental measurements. The computational and

experimental minimum fluidization velocities are 0.24m/s and 0.26 m/s with 8%

deviation. Experimental and computational pressure standard deviation also

gives the same results for minimum fluidization velocity. The computational and

experimental pressure drops are about 29 and 27 mbar respectively at the

minimum fluidization condition. The deviation is about 7%. Computational and

experimental bed expansions at the minimum fluidization condition are 8% and

10% respectively. Computational solid volume fraction fluctuation is studied to

predict bubble formation and bubble frequency.

For the given operating conditions, the model predictions are in good

agreement with the experimental measurement. The model can be the basis for

the study of flow behaviour in high temperature steam fluidized gasifiers.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 239

References

[1] V.Mathiesen, T.S., B.J. Hjertager, Prediction of gas particle flow with an

Eulerian model including a realistic particle size distribution. Power

Technology, 2000. 112: p. 34-45.

[2] Taghipour, F., Ellis, N., Wong, C., Experimental and computational study of

gas-solid fluidized bed hydrodynamics. Chemical Engineering Science,

2005. 60: p. 6857-6867.

[3] Hamzehei, M., Rahimzadeh, H, Ahmadi, G., Studies of gas velocity and

particle size effects on fluidized bed hydrodynamics with CFD modeling and

experimental investigations. Journal of Mechanics, 2010. 26.

[4] Wachem, v.B.G.M., Schouten J.C., Krishna, R., Bleek den van C. M.,

Validation of the Eulerian simulated dynamic behavior of gas-solid fluidized

bed. Chemical Engineering Science, 1999. 54: p. 2141-2149.

[5] Sahoo, A., Ch, R., Biswal, K.C., Experimental and computational study of

the bed dynamics of semi-cylindrical gas-solid fluidized bed. Canadian

Journal of Chemical Engineering, 2009. 87: p. 11-18.

[6] M.M. Kumar, E.N., CFD simulation for two-phase mixing in 2D fluidized

bed. Int J Adv Manf Technol, 2008.

[7] Jayarathna, S.A., Recommendation of a model for simulating and analysis of

the influence of particle size distribution on the simulation of bubbling

fluidized bed, in Department of Process, Energy and Environment. 2008,

Telemark University College: Prosgrunn. p. 22-46.

[8] Gidaspow, D., Multiphase Flow and Fluidization. 1994, California:

Academic Press Inc.

[9] Kunii, D., Levenspiel, O., Fluidization Engineering. Second ed. 1991,

London: Butterworth-Heinemann.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

This page intentionally left blank

www.engbookspdf.com

Advances in Fluid Mechanics IX 241

of biomass

K. Kleiva, R. K. Thapa & B. M. Halvorsen

Department of Process, Energy and Environmental Technology,

Telemark University College, Norway

Abstract

Experimental studies have been performed on a cold fluidized bed with glass and

plastic particles emulating the properties of wood and olivine particles. The

purpose of the experimental studies was to find the fluidization properties of the

particles. These experimental studies have founded the background for

conducting computational studies of multiphase flows of particles with the same

properties. The computational studies are performed in ANSYS Fluent 12.0. The

mathematical model has been verified with an experimental study to validate the

simulations. A combination of steam, olivine particles and wood chips has then

been simulated based on the results from the experimental and computational

studies. The results from the computational studies show that the minimum

fluidization velocity in the Fluent-simulations is higher than in the physical

experiments, which is expected to a certain degree because of the wider

distribution of particle sizes in the physical experiment. The pressures in the

simulations scale well with the experimental results with increasing superficial

velocity; more than one solid phase in the simulation gives a higher difference in

pressure between the simulated and experimental values.

Keywords: fluidized bed, CFD, multi-phase Eulerian model, multiple density

simulations, biomass, CHP.

1 Introduction

The behavior of biomass particles inside a fluidized bed is an essential part of the

efficiency of a gasification combined heat and power-plant (CHP-plant), and for

simulations of the behavior to be accurate, physical experiments are vital. The

computational model used in fluent is based on already published models [1].

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120221

www.engbookspdf.com

242 Advances in Fluid Mechanics IX

to be adequately high, good mixing is necessary in the fluidized bed unit. The

process itself should be able to generate enough heat so that external energy is

not required to feed the process when it is up and running. A big span in size and

density between the reacting particles and the inert particles in the bed lead to a

natural segregation of the lighter and denser particles. The biomass tends to float

on top of the bed, volatiles are released into the freeboard and the requirements

for heating the steam inside the bed might be unfulfilled, making it difficult to

maintain sufficient gasification temperature. This results in a much lower

efficiency of the gasification process than what is possible with a more uniform

mixing [2].

The adaptation of a mathematical model to a fluidized bed is already covered by

Sanoja and Ariyarathna [3]. The setup used in Fluent is presented in Table 1.

Solid-fluid drag model Syamlal-O’Brien

Solid-solid drag model Syamlal-O’Brien Symmetric

Granular viscosity Syamlal-O’Brien

Granular bulk viscosity Constant

Frictional viscosity Schaeffer

Frictional pressure Based-ktgf

Solids pressure Ma-ahmadi

Radial distribution function Ma-ahmadi

The different mathematical models are presented in this section. Multiphase

simulation solves the continuity, momentum and energy equations for each

phase. For this study, the two most important are the continuity equation and the

momentum equation as there is no transfer of heat between the phases. The

minimum fluidization velocity (umf) of a mixture characterizes its fluidization

properties and is a way to compare the experiments with the simulations [4].

A small scale fluidized bed is used in the experimental study. The bed has

pressure sensors located on the walls with a vertical distance of 10 cm between

them. A drawing of the bed setup is shown in Figure 1.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 243

and pressure sensor alignments (not drawn to scale).

The air diffuser, seen in Figure 1, in the bottom of the bed gives a uniform air

distribution at the inlet of the bed. The eight pressure nodes are connected to a

hub which again is connected to LabVIEW. The diameter of the circular bed is

8.4 cm and the height is 1.2 m.

characteristics as well as the air phase characteristics used in the Fluent

simulations are displayed in Table 2.

Mean

Density Viscosity Geldart

Phase particle size

(kg/m3) (kg/m-s) classification [5]

(µm)

Glass 182 2485 - B

Plastic 3000 964 - D

Air - 1.225 1.7894e-05 -

Three sets of experiments were run with different mixtures of glass and

plastic particles, as seen in Table 3.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

244 Advances in Fluid Mechanics IX

Table 3: Mixture ratios of glass and plastic in the G, GP-1 and GP-2

simulations.

Simulation glass plastic Total mixture of plastic

particles (l) particles (l) particles

G 2 0 0 vol% plastic 0 wt%

material properties that are of the same kind that appear in a real situation, a

steam/olivine/charred wood simulation has been performed. The characteristics

of the different elements of this simulation are presented in Table 4.

simulation.

Mean

Density Viscosity Geldart

Phase particle size

(kg/m3) (kg/m-s) classification

(µm)

Steam (800°C) - 0.29 4.1e-05 -

Olivine 400 2500 1.7894e-05 B

Charred wood 3000 400 1.7894e-05 D

The weight-percentage of the charred wood particles is set to 10 wt-%, giving a

vol-% of 38.5 of the wood particles patched uniformly with olivines in the initial

conditions in the fluidized bed simulation.

Total mixture

olivine (l) wood (cw) (l) particles

The snapshots from the simulations of the GP-1 mixture, Figure 2, show that

small bubbles appear in the bed at around 8 cm/s, showing that umf is reached.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 245

The course of the volume fraction of air as a function of time in point p3,

13.5 cm above the air-inlet, of the GP-1 mixture for a superficial velocity of 6

and 8 cm/s are shown in Figure 3 and Figure 4 respectively. Fluctuations in the

volume fraction of air are a sign of fluidization.

Figure 3 and Figure 4 indicate, as the snapshot in Figure 2 that the umf is

reached before a superficial velocity of 8 cm/s.

Volume fraction

Time step (ms)

Figure 3: Course of the facet average volume fraction of air in p3 of the bed

for the simulation of the GP-1 mixture at a superficial velocity of

6 cm/s as a function of time.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

246 Advances in Fluid Mechanics IX

Volume fraction

Time step (ms)

Figure 4: Course of the facet average volume fraction of air in p3 of the bed

for the simulation of the GP-1 mixture at a superficial velocity of 8

cm/s as a function of time.

The difference in the particle properties of the glass and plastic particles is

clearly seen in Figure 5. The plastic particles float to the top because of their

lower density and larger diameter, as also was visually apparent in the physical

experiment. The snapshot at 14 cm/s has a more uniform mixture of glass and

plastic particles, suggesting that the vertical mixing of the two particle types is

better with increased superficial velocity from umf. The expansion of the bed is

also making the distribution of plastic particles more uniform.

mixture.

The GP-2 mixture, Figure 6, has a slightly lower umf than the GP-1 mixture,

and this picture also shows bubble appearance at 6 cm/s in the top part of the

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 247

bed. This holds true until enough light particles are added to the mixture to

reduce the umf once again when the packing no longer is optimal.

Ramakers et al. [6] stated that a mixture of 10 wt-% of wood particles in a

olivine-wood mixture is an optimal mixture for uniform mixing of the particles

in a fluidized bed. This theory is arguably notable in the comparison of Figure 5

and Figure 7 which have a wt-% of 10.5 and 17.5 respectively. The gathering of

plastic particles in the top part of the bed is much higher in the GP-2 mixture in

Figure 7 at a superficial velocity of 14 cm/s, which in a gasifier would result in a

mixture.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

248 Advances in Fluid Mechanics IX

poorer energy transfer from the non-reacting olivine particles to the wood

particles. Both particle mixtures have a uniform concentration in the bottom part

of the bed, but the GP-1 mixture have a more uniform concentration overall.

The simulation of the steam/olivine/charred wood-combination has been done to

transfer the proven model to a case that is closer to a real gasifier in a CHP-plant.

A simplified theoretical value of the umf for this mixture has been calculated

using the Ergun equation [4] to be approximately 0.48 m/s. The course of the

volume fraction of steam in the lowest superficial velocity simulation of 0.24

m/s, Figure 8, shows that the umf already is superceded. The large peaks in the

volume fraction of steam are an indication of bubbles and the amplitude of the

peaks are a sign of a well fluidized bed.

Volume fraction

Time step (s)

simulation, with a superficial velocity of 0.24 m/s as a function of

time.

to achieve fluidization is increased when compared to cold air as a fluidization

medium. This is because of the lower density of the steam. The lowest simulated

velocity of 0.24 m/s gives a proper fluidized bed, as can also be seen in Figure 9.

The bubbles are comparable to the GP-1 simulation, but it is apparent that the

superficial velocity of 0.24 m/s is slightly higher than the umf for the particle-gas

mixture. The most homogenous mixing of the combination of olivine and

charred wood particles, see Figure 10, seems to be somewhere in between a

superficial velocity 0.48 m/s and 0.72 m/s. This can be stated because the area on

the top the bed with charred wood at 0.72 m/s has a higher concentration of

wood particles than the top part at 0.48 m/s. This also supports that a higher

superficial velocity to a certain points increases the buoyancy forces working on

the lighter particles in the fluidized bed. The assumed umf of slightly below 0.24

m/s along with the fact that the most uniform mixing happens in between 0.48

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 249

m/s and 0.72 m/s further supports the suggestion by Ramakers et al. [6], that a

superficial velocity of 3-4 times the umf is the ideal superficial velocity for

optimal mixing of wood particles and non-reactive particles in a fluidized bed.

4 Conclusion

The glass and plastic experiments show that adding plastic particles to the

mixture causes an increase of the minimum fluidization velocity (umf) to a certain

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

250 Advances in Fluid Mechanics IX

point, before it decreases again. The addition of larger particles causes the

packing factor of the glass particle mixture to increase to a degree, and hence the

umf to increase. When more light density and large particles are added to the mix,

the umf will start decreasing again as they have a smaller theoretical umf, the

mixture of 23 vol-% plastic was found to have a umf of 0.05 m/s, while the 33

vol-% plastic mixture was found to have a umf of 0.045 m/s. In the case of mixed

glass and plastic particles, the fluidization seemed to appear in both the top and

bottom region of the fluidized bed simultaneously.

The simulations gave a higher umf than the experiments for both mixtures of

glass and plastic particles as well as for glass particles only. This is an effect of

not having a particle distribution in the particles, only a mean diameter. The 23

vol-% plastic mixture has a simulated umf of 0.09 m/s, while a simulated

33 vol-% plastic mixture has a umf of 0.075 m/s.

The 23 vol% plastic mixture correlates to a 10.5 mass% mixture, which is

assumed to be the optimum mixture between large light density particles and

small high density particles. This mixture gives a more uniform distribution of

reacting and catalytic particles in the bed. When a 10 mass% mixture of wood

chips in a steam, olivine and charred wood mixture is simulated, it is apparent

that the uniformity of the mixture does not get better with a large superficial

velocity. Large superficial velocities increase the buoyancy effect, causing the

light particles to gather in the top of the particle region in the bed. A superficial

velocity of 3-4 times the umf seems to be the optimal conditions for a

steam/olivine/wood chip gasifier.

References

[1] P. L. N. Ramesh and M. Raajenthiren, “A Review of Some Existing Drag

Models Describing the Interaction Between the Solid-Gaseous Phases In a

CFB,” International Journal of Engineering Science and Technology,

vol. 2, pp. 1047-1051, 2010.

[2] J. X. Laihong Shen, Fredrik Niklasson, Filip Johnsson, “Biomass mixing in

a fluidized bed biomass gasifier for hydrogen production,” Chemical

Engineering Science, pp. 636-643, 2006.

[3] D. G. A. Sanoja U. Ariyarathna, “Recommendation of a Model for

Simulating & Analysis of the Influence of Particle Size Distribution on the

Simulations of Bubbling Fluidized Beds,” Master Thesis, Telemark College

University, 2008.

[4] D. Kunii and O. Levenspiel, Fluidization Engineering: Butterworth-

Heinemann, 1991.

[5] D. Geldart, “Types of Gas Fluidization,” Powder Technology, pp. 285-295,

1973.

[6] Bram J. Ramakers, Ronny de Ridder and Piet J.A.M. Kerkhof “Fluidization

behavior of wood/sand mixtures,” Drying, pp. 1337-1344, 2004.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 251

hydrostatic pressure

A. Gajewski

Bialystok University of Technology, Faculty of Civil Engineering and

Environmental Engineering, Department of Heat Engineering, Poland

Abstract

The surface-tension forces and the pressure force at a gas–liquid interface were

balanced. This was done at a differential control surface under hydrostatic

pressure. As a result, three quadratic and two linear equations were derived

representing the conditions for equilibrium at the interface. These equations have

eight roots describing eleven different equilibrium states. The most important is

the Young-Laplace formula under hydrostatic pressure which is denoted in the

spherical coordinate system. Hence, it is a quadratic equation. The first root of

this equation describes a pendant drop, drop or bubble, while the second one

circumscribes a sessile drop. There are two solutions for the uniform pressure,

where one of them is the Young-Laplace formula. It is concluded that the surface

existence depends on the pressure difference between both bulk phases. A plane

is formed when the surface-tension forces equilibrate themselves. Under uniform

pressure the interface is a sphere and its other shapes need a pressure gradient.

To form a drop or a bubble the pressure difference must be higher than its border

value. A sessile drop exists if the gauge pressure is negative, while a pendant

drop requires a positive value. A comparison with experimental results is done

for the bubble.

Keywords: Young-Laplace equation in spherical coordinates, static equilibrium,

drop, bubble.

1 Introduction

To describe the force balance on the liquid–gas interface, the Young-Laplace

equation is usually applied. The equation must be solved for each point

separately. Hence, it is computationally difficult to balance the surface-tension

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120231

www.engbookspdf.com

252 Advances in Fluid Mechanics IX

Laplace equation was derived from work involving the expansion of a soap-

bubble surface. The same formula can be also applied as a condition of static

equilibrium because the surface tension is considered simultaneously as a

surface-tension force per unit length and, following Tolman [1], either as the

quotient of the energy change with the change in surface area at constant

entropy, constant composition and constant volumes for the two phases

(measured at the surface of tension) or as the rate of free energy change with the

change in surface area at constant temperature, constant composition and

constant volumes. The theory of wetting phenomena on curved surfaces has been

developed to date by the derivation of formulas or theorems originating from an

“energetic” definition and their application in force balances. Hence, this

solution is limited to one scalar equation. Because a force may have three

components, this equation may not be applied as a condition for equilibrium in

the case of an unsymmetrical shape, nor can the shape of the surface be

described using only the force balance due to the lack of the corresponding

equations along the tangents to the surface directions.

The main goal of the current work is to develop a wetting theory of curved

gas-liquid interfacial surfaces beginning directly from the surface tension defined

as the unit force. This was expected to yield two equations of force balance for

the symmetric shapes for the droplet or bubble under hydrostatic pressure.

2 Mathematical model

The topic of discussion here is the prediction of the shape of the liquid–gas

interface under hydrostatic pressure. To do it, the forces applied on the surface at

rest are balanced. As surface forces shape the surface, they should be balanced at

the surface. While the fixed coordinate system is used, the hydrostatic pressure

creates a changing radius of curvature. Therefore, the surface with a varying

radius along the depth of the liquid must be described.

The surface presented in fig. 1 is defined in the spherical parameterisation in

which the radial distance changes along the coordinate:

f , r( ) sin cos ,r( ) sin sin ,r( ) cos (1)

The versors of the surface-coordinate system are

(2)

δ cos cos i cos sin j sin k

δ sin i cos j (3)

where is defined in fig. 3.

Moving along the tangential vector, the surface curves in the direction of the

movement. The curvature is described by the unit vector perpendicular to the

surface. It arises from the structure of the surface, which is a two-dimensional

object existing in three-dimensional space. Taking the above into account, it is

necessary to define the normal unit vector:

δ n sin cos i sin sin j cos k (4)

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 253

n

r

k

O j

i y

r

x

Figure 1: The assumed surface coordinates and the versors.

No matter how small, a fluid is a substance that deforms continuously under the

application of tangential stress. In the considered case, the shear stresses equal

zero. Therefore, only the normal stresses are sustained. As a result, the resultant

force acting on the surface is normal to the surface. The resultant pressure force

that tends to expand the surface, acts in the centre of the area. Because the force

system has to satisfy Newton’s third law, the directions and senses of the

resultant surface forces preventing expansion must be oriented out from the

surface. However, the components of the resultant surface-tension forces must be

applied at the boundary arc of the surface, and these are tangential to the surface.

The force system is shown in fig. 2. In this case, the resultant pressure force is

the active force. Therefore, the resultant of the tangential surface forces is the

reactive force. A differential surface element is defined using two arcs with radii

r and rsin (cf. fig. 2). Therefore, this surface is a curvilinear rectangle (fig.1 and

fig. 2). Hence, we have four infinitesimal surface-tension forces (fig. 2). As a

result, we obtain two resultants, one each of each pair of these forces. These

resultants equilibrate the infinitesimal pressure force.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

254 Advances in Fluid Mechanics IX

z dF

rsin dl

dF

d

dF

r dl

dl dA

d

dl

O

dF

dFp

r 2 sin d d

dF p pdA p δ n 00 (5)

cos

The surface tension forces in the component are presented

dF 01 rd cos δ 01 (6)

dF 0 1 rd cos δ 0 1 (7)

Next, the components in the direction are obtained:

dF 10 r dr 2 sin d 2 d δ 10 , (8)

dF 10 r dr 2 sin d 2 d δ 10 . (9)

The resultant force in the direction is:

d

dF r sin d cos dr sin cos 2 d δ 0 0 r sin sin d d δ n 0 0 (10)

2

and in the direction:

rd d rd d

dF 2 cos sin δ 00 2 sin sin δ n 0 0 (11)

cos 2 cos 2

The system of the resultant forces and the cross section of the investigated

surface are showed in fig. 3. Hydrostatic pressure is calculated as follows:

p g r0 r cos . (12)

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 255

dF dF

2

r/

r0

r-d r

/ 2

d

r/2

d +d

r

dFp dr

0

Figure 3: The force system resolved into meridian and normal directions.

The resultant surface tension forces and pressure force are balanced.

Respectively, their components along the meridian and normal directions to the

surface must equal zero. First we obtain the force balance in tangential direction:

dr

tan 1 dr rd

2

1 (13)

rd

and the balance of the forces in the normal direction:

dr 1

r 1

d gr r0 r cos

2

(14)

1

Substituting the relationship

r r g (15)

into Eqs. (13) and (14), we obtain a system of dimensionless equations as

follows:

dr dr

tan 1 2 tan 0

2

r d r d

dr 1 (16)

1 0

r d r ro r cos 1

2

for which the solution is the product of the left sides of the five equations:

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

256 Advances in Fluid Mechanics IX

r 0 (18)

r r cos 0

o

(19)

r ro r cos 2 sin 2 0 (20)

r

r

o

r cos 2 cos 0

2

(21)

These equations can be expressed in dimensional form as

g r0 r cos 2 r (22)

r 0 (23)

g r0 r cos 0 (24)

g r0 r cos 2 sin 2 r (25)

g r0 r cos 2 cos r 2

(26)

Formulas (17) and (22) are the Young-Laplace equation for the hydrostatic

pressure written in spherical coordinates. These formulas can be solved as a

quadratic equation with r as the variable. Hence, the shape of the surface can be

found easily.

In contrast, their solution in the Cartesian coordinate system, in which the

double surface tension is divided by the radius of curvature, requires the solution

of a second-order differential equation. It can be concluded that the surface can

exist if the resultant surface-tension force equals zero (cf. Eq. (24)), but it cannot

be formed if the hydrostatic pressure is equal to zero. The last conclusion arises

from the analysis of Eqs. (22), (24), (25) and (26) and is expressed implicitly by

Eq. (23).

As the shapes of the liquid–gas interfaces are presented by the solutions of Eqs.

(17)–(21), we can distinguish four kinds of surface equilibrium. The first is

described the Young-Laplace formula, i.e., Eq. (17). The second is the flat

surface, expressed by Eq. (19), with a lack of a surface as its particular case (cf.

Eq. (18)). The third and fourth kinds of equilibria have not been connected with

any real situation so far. Therefore, they can be only the solutions of the force

balance without any physical meaning.

Although the analysis is made in the surface-coordinate system, the graphs

are plotted in Cartesian coordinates:

x r sin (27)

z r cos (28)

where the variables are defined in fig. 1. Because these solutions have

axisymmetric shapes, they are plotted as cross-sections in the x+z+ plane.

The solution of Eq. (17), i.e., the Young-Laplace formula with hydrostatic

pressure in spherical coordinates, has two roots:

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 257

r1 (29)

2 cos

ro ro2 8 cos

r2 (30)

2 cos

Because these solutions have discontinuities at , the limits of the

functions at this point must be calculated. To do so, we represent the square roots

in Eqs. (29) and (30) as Taylor series. If r0 r02 1 , the limits of the roots r1

and r2 are calculated as follows:

2

lim r1 (31)

2 ro

lim r2 (32)

2

lim r1 , (33)

2

2

lim r2 (34)

2 ro

Hence, in the case of a positive hydraulic head, r0 , the root r1 is a

continuous function. If the hydraulic head has a negative value, the solution for

r 2 has no discontinuity. We can summarise that the hydraulic head is the

asymptote for some solutions. Note that r1 and r2 can generally be complex

numbers. To return a real number for the square root over the interval

r0 cannot be less than 2 2 . Over the domain cosine is less than

zero, and so the discriminant has a positive value. The solutions of these

equations for this positive border values are presented in fig. 4. The graphs for

r1 and for r 2 over the interval show the shapes of the interfaces

when the pressure inside is higher than their surroundings. The plots for r2 over

the interval present the surface limited in volume with internal

pressure lower than the surroundings.

In the case of higher gauge pressure, the drop has no tangent point with the

upper layer, but its shape is almost spherical because the vertical diameter is

slightly longer than the horizontal one. These diameters will be equal if

r0 55 .2 , and so the drop will then be a sphere.

The sample solution for 0 r0 2 2 is presented in fig. 4b). The intervals in

which the real roots of eqs. (29) and (30) can be found are noted in this figure.

The results for r1 model the case of a pendant drop that is still connected to the

solid phase. We can conclude that the drop will form if the dimensionless gauge-

pressure value reaches 2 2 . Although gaps in the plots are observed for r2 in

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

258 Advances in Fluid Mechanics IX

the domain , this does not indicate a lack of the surface at the gap

period in this case; rather, it only means the lack of a state of the equilibrium for

the surface with the assumed force system. The surface could thus accelerate, or

other forces could be introduced to the analysis. A different situation is observed

in the interval . Because cosine is less than zero here, the

discriminant in Eq. (30) is a positive value. Thus, there is no gap for r2 within

this interval.

5 5

z+ z+

4 4

3 hydraulic 3

head

hydraulic

2 2

head

1 1

0 r1+ 0 r 1+

r 2+ r2+

-1 -1

+ +

-3 -2 -1 0 1 2 x 3 -3 -2 -1 0 1 2 x 3

a) b)

Figure 4: The plots of eqs. (29) and (30) for: a) r0 2 2 , b) r0 2 .

The plots for 2 2 r0 0 are the mirror image of the plots for their

modulus values. There occurs the shift between solutions for Eqs. (29) and (30).

The second root describes the sessile drop well. Therefore, the plots for r0 2

are the mirror image of the plots in fig. 4b), where the hydraulic head line is the

axis of symmetry.

The flat surface is the solution of Eq. (19):

r

r4 o (35)

cos

The height of the surface or the depth of the liquid is equal to hydraulic head,

r0 . As the right side of Eq. (19) is zero, there is no resultant surface-tension

force. However, this does not mean that these forces are not present but rather

that the surface-tension forces achieve equilibrium on the plane and only their

resultant does not exist.

The most extreme state is the lack of a surface if the hydraulic head equals

zero, which is expressed by Eq. (20). This is an important finding of this work.

The surface can exist if the surface-tension forces balance each other, but the

surface cannot be shaped if there is no pressure difference between two sides of

the surface (i.e., between the two bulk phases). This conclusion is confirmed by

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 259

the limit of the function calculated by Eq. (32). A liquid–gas surface will not

reach the level of the hydraulic head if the surface-tension forces do not achieve

equilibrium. These situations are presented in figs. 4, 5.

5 5

z+ z+

4 4

3 3 hydraulic

head

2 hydraulic 2

head

1 1

r 5+ r7+

r 8+

+

0 r6 0

-1 -1

+ +

-3 -2 -1 0 1 2 x 3 -3 -2 -1 0 1 2x 3

a) b)

b) The graphs of Eqs. (38) and (39) for r0 2 2 .

This solution has not been connected with any experimental observations so far,

and its properties have not been described in detail.

The third equilibrium state, defined by Eq. (20), has two roots:

ro ro2 8 sin 2 cos

r5 (36)

2 cos

ro ro2 8 sin 2 cos

r6 (37)

2 cos

The graphs of Eq. (36), plotted in fig. 5a) have shapes similar to a horn torus.

The plots of Eq. (37) are tangent to the line of the hydraulic head because sine

for in Eq. (20) is equal to zero. For that reason, this equation is

decomposed into two: (18) and (19) (i.e., the lack of a surface and a flat surface,

respectively) in this point. The solution for the plane is plotted in fig. 5a) at

which makes this solution different from the cases described in Sections

2.3.1 and 2.3.4. The depth of the lower layer (i.e., the difference between the

hydraulic head and the vertical coordinate of the lower surface) is tangent to the

upper plate (the solution for and it increases up to its maximum

value (for = 0.9553) then approaches the hydraulic-head line. The height of the

upper layer (the solution for rises from 0 (for = ) to its maximum

(for = 2.186) and then approaches the hydraulic-head line.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

260 Advances in Fluid Mechanics IX

The fourth equilibrium state, given by Eq. (21), has two roots:

ro ro2 8 cos 3

r7 (38)

2 cos

ro ro2 8 cos 3

r8 (39)

2 cos

The seventh root of the general force balance given by Eq. (38) has the shape

of a thin drop suspended within another one, cf. fig. 5b). The upper layer (the

solution for is largest for = 0, and then it decreases and approaches

the hydraulic-head line. The lower layer (the solution for is deepest

for 0, and it monotonically approaches the hydraulic-head line.

Eq. (14) under uniform pressure takes the form

dr 1

r 1 ,

d pr

2

(40)

1

while eq. (13) does not change. There are two solutions of eqs. (13) and (40):

2

p , (41)

r

2 2

p cos . (42)

r

Formula (41) is well known Young-Laplace equation and describes bubble

under uniform pressure that shape is spherical. Although, it is common known

result, for the first time this shape was derived from the complete force balance

directly. Eq. (42) seems to be only mathematical solution that has not been

observed experimentally, which cross-section is two equal bubbles where second

bubble is attached under the first one.

It was proven that it is possible to obtain a complete system of equations for the

static equilibrium of a surface under hydrostatic pressure for all fluids that do not

sustain shear stresses when at rest. To predict the shape of this surface, a new

differential surface with a varying radial distance along the meridian direction

was defined. The pressure force and surface-tension forces were balanced on this

surface. In the method presented herein, it was possible to balance the surface

forces and the surface-tension forces on the control surface. This had not

previously been possible, and it represents the major finding of this work. As a

result, a system of two equations was obtained: one each in the radial and

tangential directions. Their solutions were decomposed into three quadratic and

two linear equations with eight roots. Due to the existence of discontinuities,

they can describe eleven different situations. The first solution is the Young-

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 261

Laplace equation with the interface under hydrostatic pressure. To obtain the

results in the traditional way, a second-order differential equation must be

solved. In the achieved method, a simple quadratic equation is instead solved,

which makes the calculations quite simple. The linear solutions are the flat

surface and the lack of a surface. Although they are both trivial, their presence

proves the generality of the developed model. Furthermore, two new solutions

were obtained that seem to be unconnected with any physical situation.

3.1 Conclusions

We drew the following conclusions for a surface under pressure and surface-

tension forces only:

A surface under uniform pressure is a sphere.

A plane is formed if the surface-tension forces equilibrate themselves.

Other shapes occur if a pressure gradient exists and the surface-tension

forces equilibrate pressure force.

A surface cannot exist without a pressure difference across both of its

sides.

The pressure forces are the active forces, whereas the surface-tension

forces are the reactive forces, which only act to curve the surface.

To form a closed interface (e.g., a drop or a bubble), the interior

pressure cannot be less than its border value; i.e., the modulus of the

hydraulic head must be a factor of 2 2 or more times higher than the

capillary constant.

Acknowledgement

This scientific research was financed from the resources of the National Centre

for Science.

References

[1] R.C. Tolman, Consideration of the Gibbs Theory of Surface Tension, The

Journal of Chemical Physics, Vol. 16, No. 8, 1948, 758-774.

[2] Thoroddsen S.T., Etoh T.G., Takehara K., High-Speed Imaging of Drops

and Bubbles, Annu. Rev. Fluid Mech. 2008, 40:257-85.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

This page intentionally left blank

www.engbookspdf.com

Advances in Fluid Mechanics IX 263

slug flow transition modelling

R. Denefle1 , S. Mimouni1, J.-P. Caltagirone2 & S. Vincent1

1

EDF R&D, France

2

I2M, Université Bordeaux 1, UMR 5295, France

Abstract

This paper aims to assess a conservative interface locating method based on level-

set adapted to two fluid model for two phase flow. The objective of the work is

twofold: first to check out the prediction of the model on bubble coalescence cases

and then to introduce a first description of the multifield hybrid approach for two-

phase flow modelling. The interface locating method is part of a model dedicated

to the simulation of large and distorted bubbles. At different liquid viscosities

and densities, the model yields reasonable predictions of terminal velocities and

shapes for rising bubble experiments. The overall method relies also on an existing

building block, consisting of a set of averaged models dedicated to dispersed

bubbles, which has already been validated and has given a reasonable agreement

with experimental data in cases where the spherical shape assumption is still valid

for the dispersed phase. As a consequence, we present the first step toward a

new approach consisting of modelling the small bubbles as a dispersed field and

simulating the large and distorted ones.

The main outcome is the simulation of bubble coalescence where the distortion

of the interface during the coalescence phenomena is followed. The capability to

simulate correctly coalescence phenomena is an important point of the modelling

of slug flows with interface locating methods.

Keywords: multifield model, surface tension, interface sharpening, bubble rising,

coalescence.

1 Introduction

Two-phase flows are featuring many industrial applications such as nuclear power

plants, heat exchangers and chemical reactors. Based on the interface structure,

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120241

www.engbookspdf.com

264 Advances in Fluid Mechanics IX

main groups. The first group would be separated flows such as annular, film or

stratified flows, the second the dispersed flows such as bubbly, droplet or particle

flows, and the last group would be the flows such as bubbly annular, churn

turbulent or slug flow. Although these regimes have been experimentally confirmed

since a few decades [1], their numerical simulation is still challenging and an

universal model remains to be established.

There are several approaches for two-phase flow Eulerian modelling, that may

be classified by the successive numerical choices, such as the number of fields,

the use of space or time averages, the filtering of some space scales, and the

way the interfaces are dealt with. As described by Bestion [2], the capability of

these approaches to model the different flow regimes are far from being similar.

The RANS approach, which is today of widespread use in CFD for industrial

applications presents the drawback of filtering every intermittency scales. This

means that besides this approach can deals with all flow regimes, it requires a

large amount of work on the closure laws models, currently limited to dispersed

and separated flows. The DNS and LES without interface filtering are nowadays

too expensive to treat high Reynolds number flows such as churn. According to

Bestion, an hybrid method so called Hybrid LES, both filtering and simulating

interfaces may be a promising way to describe all flow regimes at a reasonable

CPU cost without filtering the two-phase structures as RANS approach does.

We present here the first steps building the Hybrid LES method. We propose to

treat statistically the interfaces small enough to be considered as spherical, and to

simulate the distorted interfaces. This naturally leads to a filter scale imposed by

the flow conditions.

This method is implemented in the multifield code NEPTUNE CFD. The

modeling of dispersed phase flows has already been validated [3] and has given

a reasonable agreement with experimental data in cases where the spherical shape

assumption is still valid for the dispersed phase. As a consequence, we concentrate

here our work on the simulation of large and deformed bubbles. A first step toward

the Hybrid LES method is to build a set of equations to model the large interface

field. Numerical model and its application to bubble rising are summarized here

and can be found in [4] for more details.

After proving the efficiency of the method on well-known bubble rising cases,

we will test the model on two-bubble coalescence experimental cases. This will

comfort the extension of the interface locating model toward slug flows studying.

physics. Despite several theoretical [5], experimental [6] and over these recent

years numerical [7, 8] studies, this paradigm is still a topic of major interest.

Most of the current techniques applied to interface flows such as large

bubble rising have been developed taking into account two major aspects:

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 265

discontinuous fluid properties at the interface.

Several categories of interface locating exist. The first one is Lagrangian grid

methods [9] where the background grid is following the interface. The main

limitation is that this method cannot track surfaces that break apart or intersect.

The second category is the front tracking method [10]. The motion equation for the

flow field is solved on a fixed grid and the interface position is tracked explicitly

by markers distributed evenly on the interface. Difficulty generally comes from

the repartition of the markers which should ideally be kept equidistant, especially

for coalescence of interfaces applications. The third category of these methods

consists in capturing the interface using various volume functions defined on

the fixed grid used to solve the governing equations of the flow. The volume

of fluid [11] method and level-set [12] approaches are parts of this group. VOF

methods are conservative, but interface properties such as normal and curvature

are difficult to calculate accurately. Level-set methods automatically deal with

topological changes but are not conservative. Several techniques have been

developed to ensure volume conservation and improve the locating method, such

as a combination of Level-set and VOF method [13], or the augmented Lagrangian

method coupled to VOF techniques [14].

Olsson and Kreiss [15] introduced a level set method with an artificial

compression step performed after the advection of the level set function to ensure

that the thickness of the transition layer is preserved.

In our work, the Eulerian approach is use. This approach of the two-fluid model

allows the direct use of the volume fraction as an interface function. Thus, the

equivalent to the level-set function is self-transported by the resolution of the

momentum balance equation. To keep a good location of the interface over time,

the same artificial compression step as Olsson’s is used.

Concerning coalescence, only few experiments present results on two-bubble

coalescence such as Brereton and Korotney’s experiment [16]. Numerical works

on this phenomena usually follow interface treatment methods and can be found

with marker approach, Singh and Shyy [17], and VOF approach, van Sint

Annaland et al. [18].

3.1 Governing equations

information can be found in Ishii and Hibiki [19]. The balance equations for

isothermal two-fluid model can be written as mass and momentum balances:

∂t (αk ρk ) + ∂xi (αk ρk Uk,i ) = Γk with Γk = 0 (1)

k

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

266 Advances in Fluid Mechanics IX

∂t (αk ρk Uk,i ) + ∂xj (αk ρk Uk,i Uk,j ) = ∂xj (αk τk,ij ) − αk ∂xi (P )

(2)

+αk ρk gi + Ik,i

where ρk , Uk , αk , Γk and τk,ij , P and Ik,i denote respectively the density, the

mean velocity, the volumetric fraction, the mass transfer and the total stress tensor

including laminar and turbulent contribution, the mean pressure and the interfacial

momentum exchange between phases. The interfacial momentum exchange is

constituted here by two interfacial forces : surface tension force and drag force.

This later appears in momentum balance equations due to averaging. Its local

sum over phases is equal to zero and depends on the relative velocity between

the phases.

→

− α1 α2 (−

→

v2 − −→

v1 )(α1 ρ1 + α2 ρ2 )

F drag,1 = (3)

τ

As we consider in the paper a local separated flow in the two fluid model, the

role of the drag force is to enforce a no slip condition and so equality of the phase

velocities at the bubble interface. The closure law, eqn. (3), is applied with a very

short relaxation time τ to achieve large interface drag (τ = ∆t/100).

therefore the surface tension, eqn. (4), has to be taken into account in the model.

From an Eulerian point of view, a surface force has to be implemented in volume:

−

→

F sta = σκ−

→

n (4)

where σ, κ and − →

n denote respectively the surface tension, the curvature and the

normal vector. The curvature κ is defined as the divergence of the unit normal

vector at the interface. A color function c locating the interface, such as a level

set function, is a descriptor of this normal vector. The unit normal vector and the

curvature can be calculated with the color function, obtain by diffusion of .the

volume fraction. Here it becomes important that the thickness of the interface is

kept constant so that the color function gives a good approximation of the local

curvature. The Continuum Surface Force (CSF) method described by Brackbill

et al. [20] allows the reformulation of the surface tension into an equivalent volume

force, eqn. (5), that can be added to the momentum balance equation. In the two-

fluid model, this force is split between the two phases occupying the cell, since

two momentum equations are solved. The average model is here the volume one.

→k

− ∇c

F st = αk σ ∇. ∇cdV (5)

∇c

pay some attention to the interface thickness. Olsson and Kreiss [15] coupled an

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 267

The interface sharpening method consists in resolving the following eqn. (6) on a

non physical time step between two physical steps to ensure the interface thickness

to be kept constant.

∂τ (αk ) + ∇ (αk (1 − αk )−

→

n ) = ε∆αk (6)

The parameter ε controls the final interface width at convergence and is here

chosen so that the interface thickness is 2 cells. The non-physical time ∆τ is

chosen to ensure CFL and Fourier numbers under 0.5 and to minimize the number

of iteration leading to a steady interface width: ε = ∆x/2 and ∆τ = ∆x/32.

4.1 Experiment description

against the experiments of Raymond and Rosant [6]. Bubbles are released in

still water at time t = 0 s. The reference pressure is the atmospheric pressure

1013 hP a. The fluid viscosity and density (blend of water and glycerol) are

given in table 1. The final bubble velocity and shape (ratio between height h and

width w of the bubble) are studied as functions of the bubble initial diameter. An

axisymmetric geometry is used, reproducing the experiment situation of bubbles

rising in a pipe. The air density and viscosity are respectively 1.29 kg.m−3 and

1.84 ∗ 10−5 P a.s. A Dirichlet’s condition on the pressure field is imposed at the

top boundary.

inside cross section and 0.5 m height. Wall confinement effects on the bubbles

can be considered as negligible. As modelling such a huge domain will cost a

lot of calculation time, the aim of this analysis is to find the smallest possible

computational domain in which walls have negligible effect on the terminal

velocity and shape of the simulated bubbles. Radius from 2 to 6 times the bubble

Table 1: Fluid properties for the Raymond and Rosant’s experiment [6].

S1 0.687 1250 0.063

S3 0.242 1230 0.063

S5 0.0733 1205 0.064

S6 0.0422 1190 0.064

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

268 Advances in Fluid Mechanics IX

Figure 1: Effect of computational domain size on the bubble shape for an initially

8 mm diameter spherical bubble in the S5 experiment (see table 1).

diameter were tested for the simulation of the experimental case S5 (see table 1)

with an 8 mm diameter bubble. The fig. 1 presents the terminal bubble shape for

the tested computational domain size. The terminal velocity can be considered as

independent from the domain size D when D ≤ 4 Db .

A grid-independent test was carried out on the axisymmetric rise of a single bubble

in a liquid on six different meshes. According to the previous sensitivity analysis

on the domain size, the domain is chosen to have a four bubble diameter radius,

and twelve bubble diameter height. The six grids are M2 (h = Dbubble /∆x = 15),

M3 (h = 20), M4 (h = 25), M5 (h = 30), M6 (h = 35) and M7 (h = 40).

Figure 2: Effect of the mesh refinement on the bubble rising velocity for an initially

8 mm diameter spherical bubble in the S3 experiment. D is the domain

radius given here in number of bubble diameter.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 269

Fig. 2 shows that the mesh M5 provides a nearly grid-independent solution. This

grid will be used in all other runs.

Fig. 3 and 4 present the evolution of terminal velocities and shapes versus the initial

bubble diameter for the 4 experiments S1, S3, S5 and S6. The predicted velocities

are found to be in a good agreement with the experimental data, the deviation being

lower than 10%. Concerning the bubble aspect ratio, the deviation is lower than

10% for experiment S1 and S3, lower than 23% for experiment S5 and lower than

30% for S6. For the most distorted cases (experiment case S6 and bubble diameters

greater than 6 mm), part of the gap between measures and numerical prediction

could be explained by the difficulty to reproduce the exact initial conditions. In the

simulation, the bubble is initiated as a still sphere whereas in the experiment the

bubble may not be spherical at time t = 0 s, especially for experiments where the

bubble is large and the fluid less viscous.

(symbols) versus the equivalent diameter for various liquid properties.

5 Bubble coalescence

5.1 Experiment description

experimental observations for bubble rising cases. For the simulation of slug

flows, we need to be able to describe large bubble coalescence phenomena. The

experiments of Brereton and Korotney [16] provide a good qualitative description

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

270 Advances in Fluid Mechanics IX

Figure 4: Predicted bubble aspect ratio (solid lines) compared to experimental data

(symbols) versus the equivalent diameter for various liquid properties

(see table 1).

of the coalescence of two rising bubbles. Four cases are calculated: two cases

of coaxial coalescence and two cases of oblique coalescence, each type of

coalescence being treated in two dynamic conditions, described by Reynolds,

Morton and Eötvos numbers: case (a), (Re = 43, M o = 2.10−4 , E ö = 16) and

case (b), (Re = 1, M o = 30, E ö = 16). The two initial bubbles have the same

volume.

5.2 Results

a rectangular 3D box of size 0.04 × 0.04 × 0.08 m3 , the origin is placed at

the center of the bottom face. The initial bubble have a radius of 5 mm and

are placed at positions {(0,0,0.025) (0,0.08,0.125)} and {(0,0,0.025) (0,0,0.125)}

respectively for the oblique and the coaxial coalescence experiments. Regarding

the discretization, the space scale is ∆x = 5.10−3 m for the case (a) and

∆x = 2.5.10−3 m for the case (b), the fact that the most viscous case need

the finest grid is due to the difficulty to reproduce the sharp tail of the second

bubble observed on fig. 6 at 0.09 and 0.12 s. The time scale of the coalescence

is well predicted by the numerical simulations for all cases, as the time between

each experimental photograph is approximatively 30 ms. Despite the difficulty of

the phenomena, the shapes of both leading and trailing bubbles are in excellent

agreement with the experimental observations.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 271

and Korotney [16] and numerical predictions for coaxial and oblique

coalescence of two bubbles: Re = 43, M o = 2.10−4 , E ö = 16.

and Korotney [16] and numerical predictions for coaxial and oblique

coalescence of two bubbles: Re = 1, M o = 30, E ö = 16.

6 Conclusion

The dynamics of a deformed bubble has been studied using a two-fluid model

adapted to cases with located interfaces. Experimental validations have been

carried out. Quantitatively for bubble rising, the predicted terminal shape and

velocity of bubbles are in reasonable agreement with the experiments of Raymond

and Rosant [6]. Qualitatively for two-bubble coalescence, the coalescence time

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

272 Advances in Fluid Mechanics IX

scale and bubble behavior are found to be quite similar to Brereton and Korotney’s

experiments [16]. The model dedicated to large distorted bubbles was successfully

applied to coalescence cases in different liquid conditions. This is comforting us in

the way of an hybrid three field model to address regime transitions such as bubbly

to slug flow.

Acknowledgements

This work has been achieved in the framework of the NEPTUNE project,

financially supported by CEA (Commissariat à l’Energie Atomique et aux

Energies Alternatives), EDF, IRSN (Institut de Radioprotection et de Sûreté

Nucléaire) and AREVA-NP, in collaboration with I2M laboratory (Bordeaux,

France).

References

[1] Weisman, J. and Kang, S.Y., Flow pattern transitions in vertical and upwardly

inclined lines. International Journal of Multiphase Flow, 7, pp. 271–291,

1981.

[2] Bestion, D., Applicability of two-phase CFD to nuclear reactor thermalhy-

draulics and elaboration of best practice guidelines. Nuclear Engineering and

Design, In Press, p. Corrected proof, 2011.

[3] Mimouni, S., Archambeau, F., Boucker, M., Laviéville, J. and Morel, C.,

A second order turbulence model based on a reynolds stress approach for

two-phase boiling flow and application to fuel assembly analysis. Nuclear

Engineering and Design, 240, pp. 2225–2232, 2009.

[4] Denefle, R., Mimouni, S., Caltagirone, J.P. and Vincent, S., Multifield hybrid

LES approach for two-phase flow modelling - part 1: Adiabatic bubble rising.

Work in progress.

[5] Clift, R., Grace, J.R. and Weber, M.E., Bubbles, drops, and particles.

Academic Press, New York, 1978.

[6] Raymond, F. and Rosant, J.M., A numerical and experimental study of

the terminal velocity and shape of bubbles in viscous liquids. Chemical

Engineering Science, 55, pp. 943–955, 2000.

[7] Hua, J. and Lou, J., Numerical simulation of bubble rising in viscous liquid.

Journal of Computational Physics, 222, pp. 769–795, 2007.

[8] Koebe, M., Bothe, D. and Warnecke, H.J., Direct simulation of air

bubbles in water/glycerol mixtures : shapes and velocity fields. Proceedings

of FEDSM’03: 4th ASME JSME Joint Fluids Engineering Conference,

Honolulu, Hawaii, USA, 2003.

[9] Hyman, J.M., Numerical methods for tracking interfaces. Physica D, 12, pp.

396–407, 1984.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 273

[10] Unverdi, S.O. and Tryggvason, G., A front-tracking method for viscous,

incompressible, multi-fluid flows. Journal of Computational Physics, 100,

pp. 25–37, 1992.

[11] Hirt, C.W. and Nichols, B.D., Volume of fluid (VOF) method for the

dynamics of free boundaries. Journal of Computational Physics, 39, pp. 201–

225, 1981.

[12] Osher, S. and Sethian, J.A., Fronts propagating with curvature-dependent

speed: Algorithms based on Hamilton-Jacobi formulations. Journal of

Computational Physics, 79, pp. 12–49, 1988.

[13] Sussman, M. and Puckett, E., A coupled level set and volume-of-fluid

method for computing 3D and axisymmetric incompressible two-phase

flows. Journal of Computational Physics, 162, pp. 301–337, 2000.

[14] Vincent, S., Caltagirone, J.P., Lubin, P. and Randrianarivelo, T.N., An

adaptive augmented lagrangian method for three-dimensional multimaterial

flows. Computers and Fluids, 33, pp. 1273–1289, 2004.

[15] Olsson, E. and Kreiss, G., A conservative level set method for two phase flow.

Journal of Computational Physics, 210, pp. 225–246, 2005.

[16] Brereton, G. and Korotney, D., Coaxial and oblique coalescence of two

rising bubbles. Proceedings of AMD-Vol. 119: The ASME Applied Mechanics

Conference, Columbus, Ohio, USA, 1991.

[17] Singh, R. and Shyy, W., Three-dimensional adaptive grid computation with

conservative, marker-based tracking for interfacial fluid dynamics. 44th

Aerospace Science Meeting, Reno MV, USA, 2006.

[18] van Sint Annaland, M., Deen, N. and Kuipers, J., Numerical simulation of

gas bubbles behaviour using a three-dimensional volume of fluid method.

Chemical Engineering Science, 60, pp. 2999–3011, 2005.

[19] Ishii, M. and Hibiki, T., Thermo-fluid Dynamics of Two Phase Flow.

Springer-Verlag, 2006.

[20] Brackbill, J., Kothe, D. and Zemach, C., A continuum method for modeling

surface tension. Journal of Computational Physics, 100, pp. 335–354, 1992.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

This page intentionally left blank

www.engbookspdf.com

Advances in Fluid Mechanics IX 275

bubble column simulation approach by way

of statistical bubble micro-flow modelling

W. Coetzee1 , R. L. J. Coetzer2 & R. Rawatlal1

1

Department of Chemical Engineering, University of Cape Town,

South Africa

2

Sasol Technology Research and Development, South Africa

Abstract

Bubble columns are extensively used in the chemical process industry,

however their hydrodynamics are computationally expensive to simulate, severely

restricting design and optimisation studies. A novel discrete-phase bubble column

modelling approach is proposed where the flow-field in the immediate vicinity

of individual bubbles is predicted from the Reynolds number via an algebraic

bubble flow model, i.e. the Bubble Cell Model (BCM). This contrasts to the

traditional Euler–Lagrange approach, where the flow structure is resolved at a

larger scale using the Navier–Stokes equations. The construction of the BCM

involves generating the micro-flow fields through solving the analogous case of

the flow over a fixed sphere with the Navier–Stokes equations at 22 equally spaced

Reynolds numbers in the operating range of Re ≤ 270. The model construction

then occurs in two stages, i.e. stage one: the fitting of each velocity vector field

at the discrete Reynolds numbers, and stage two: the cross correlation of the

stage one model parameters with respect to Reynolds number. For stage one, a

hybrid between analytical and statistical models was found to provide accurate fits,

resulting in R2 values ranging from 0.9968 to 0.9999. The Reynolds dependence

of the parameters was found to be described through simple polynomial and

exponential models in the second stage construction, producing a model which

generates the velocity vector field around a bubble for a given Reynolds number

as the result of a single algebraic evaluation. The integration of the BCM with a

Eularian macro fluid model has also been investigated for a single bubble pilot

test case.

Keywords: bubble flow, Euler–Lagrange, cell model.

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120251

www.engbookspdf.com

276 Advances in Fluid Mechanics IX

1 Introduction

Bubble column simulation requires the resolution of complex hydrodynamic

interactions between the gas and liquid phases. This can be modelled with the

Navier–Stokes equations, however, the solving of the Navier–Stokes equations

at bubble scale, e.g. Volume-Of-Fluid method, is impractical with regards to

computational expense when considering any more than a few bubbles [1]. The

momentum source generated by the bubbles on the continuum fluid phase is

subsequently modelled through experimentally correlated force models i.e. drag,

lift and added mass, which is used in coupling the gas phase to the continuum [2].

The gas phase can then either be tracked as individual bubbles, i.e. the Euler–

Lagrange approach, or approximated as a continuum i.e. the Euler-Euler approach.

These approaches enable larger grid sizes and make the simulation of entire bubble

columns possible. However, the bubble scale flow detail is severely reduced and

computational efficiency remains a problem [3]. Furthermore, the question of

which physical effects to include in the coupling model, is still under debate [4].

It follows that the choice of bubble column simulation approach depends on the

level of flow detail required and the size of the geometry, as illustrated in Figure 1.

The computational expense required for large sized problems can subsequently

inhibit the choice of smaller scale models.

In an attempt to address these issues, relating to multiphase flow simulation in

general, it has been proposed by the authors to substitute the flow-fields in the

direct vicinity of the bubbles in a Euler–Lagrangian framework, with an algebraic

micro-flow model correlated to the Navier–Stokes solution for the appropriate

flow situations. The algebraic flow model will be referred to as the Bubble Cell

Model (BCM). The computationally expensive regions around individual bubbles

are therefore accounted with the BCM, which can be rapidly evaluated, in contrast

to the computationally expensive partial differential Navier–Stokes equations. In

addition, the BCM can be used to determine the force acting on the bubble

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 277

in the equation of motion and allows the bubble scale flow information to be

approximated.

The aim of the BCM is to provide a model of the velocity vector field, uM ,

with respect to spatial position, x, around a bubble as a function of Reynolds

number, Re = ρDu µ

rel

(where ρ, D, urel and µ represent the fluid density, bubble

diameter, bubble relative velocity and fluid viscosity respectively). The solution

to the Navier–Stokes equations for the flow over a fixed sphere with appropriate

boundary conditions (as discussed in Section 2.1), is regarded as the “true velocity

field”, utrue , with respect to the construction of the BCM. To approximate utrue

as a function of Reynolds number, a two stage fitting strategy has been devised.

Stage one involves fitting the velocity vector fields at discrete Reynolds numbers,

whilst stage two aims to cross correlate the parameters of the stage one model w.r.t.

Reynolds number. This strategy is schematically presented in Figure 2.

To set up the case, the boundary condition at the bubble surface and the bubble

shape is considered:

• If the liquid is pure enough, it is possible for it to slip along the surface of

the bubbles (free slip condition), in contrast to the flow past rigid bodies,

where slip will not occur (no-slip condition). This very often makes the

flow unseparated in circumstances where the flow around a solid body of

similar shape would be separated (i.e. the fluid flow becomes detached from

the surface of the object at higher Reynolds numbers) or even turbulent

[5]. Consequently, the boundary condition for a contaminant free system,

Figure 2: BCM two stage fitting strategy, where f1 , α, f2 and ζ represent the

overall stage one model, stage one parameters, stage two model and stage

two parameters respectively.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

278 Advances in Fluid Mechanics IX

velocity, is a zero-shear-stress one rather than a no-slip one [6].

• The shape of the bubble directly influences the local hydrodynamics and can

be transient during bubble motion. According to the research of Tomiyama

et al. [7], bubble motion, shape and velocity are remarkably sensitive

to initial shape deformation. The researchers subsequently proposed a

steady state drag model which is dependent on bubble shape through the

2

dimensionless Eötvös number (E ö = ∆ρgD σ , with ∆ρ, g and σ representing

the density difference, gravity vector and surface tension respectively). It has

become apparent that bubble shape is an important parameter which adds

new degrees of freedom to an already complex problem.

For the development of the BCM approach, we restrict the focus onto spherical

bubbles in a contaminant free system. It follows from experimental data [8], that

the spherical assumption is valid for Re ≤ 270 for an air-water system, whilst the

clean system will be accounted for with an zero-shear-stress boundary condition

on the bubble surface. Furthermore, to allow for faster development of the concept,

the work will be carried out in two dimensions. It should be noted that the approach

could be extended to different shaped bubbles occurring in heterogeneous bubble

columns through parametrization of the E ö number, higher Reynolds numbers and

three dimensions.

The Navier–Stokes equations are solved with the Finite-Volume method, using

the SIMPLE pressure-velocity coupling scheme [9] and 2nd order UPWIND

spatial discretization in the Fluent 12 solver (FLUENT INC, USA). The mesh

is created according to the geometry depicted in Figure 3, and its resolution

optimized according to the convergence of the drag force on the bubble surface.

CFD simulation. The geometry is set up with respect to the radius (R) of

the bubble and the ratios chosen to allow flow dynamics to sufficiently

develop across the domain.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 279

This resulted in 164,983 data points in the BCM region, which was chosen as 5

times the bubble radius. The drag coefficients for the Reynolds numbers evaluated,

corresponded to the experimental data for bubbles in pure systems [10].

Analytical solutions provide a good starting point to the development of the BCM,

since they capture useful features of the flow structure. Solutions are possible

through simplifying assumptions, such as either creeping flow (i.e. zero advective

terms), or potential flow (i.e. zero viscous terms), making analytical evaluation

of the steady state, incompressible Navier–Stokes equations (eqn. (1) and (2))

possible. Here u, p and S refer to the fluid velocity vector, fluid pressure and the

body force vector respectively.

∇·u= 0 (1)

ρ(u · ∇u) = −∇p + µ∇2 u + S

(2)

advective terms viscous terms body f orces

[11], by assuming axisymmetric flow and writing the Navier–Stokes equations

in terms of the Stokes stream function ψ (the two velocity components are

expressed as derivatives of ψ such that continuity is automatically satisfied) in

spherical coordinates ( where r and θ correspond to the radial and angular spatial

components, centred around the centre of volume of the bubble), simplifying the

problem considerably. Following the solution strategy by Slattery [12], the stream

function is proposed as a fourth order polynomial whose coefficients are obtained

through the substitution of the boundary conditions. Applying the zero-shear-stress

boundary conditions, eqn. (3)–(5), the stream function is obtained, which upon

differentiation with respect to θ and r leads to radial and angular velocities, ur

and uθ , which are subscripted for the creeping flow solution as ur,CF and uθ,CF

in eqn. (6) and (7). The free stream velocity here, u∞ , corresponds to the relative

bubble velocity urel .

ur|r=R = 0 (4)

duθ|r=R

=0 (5)

dr

3

3 R 1 R

ur,CF (r, θ) = u∞ 1 − − cos(θ) (6)

4 r 4 r

3

3 R 1 R

uθ,CF (r, θ) = u∞ 1 − + sin(θ) (7)

8 r 8 r

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

280 Advances in Fluid Mechanics IX

Although zero viscosity fluids do not actually exist, the assumption (valid at

Re → ∞) provides a reasonably good description of the velocity profile, except

near the body and beyond the line of flow separation [13]. The potential flow

solution is derived by defining velocity potential Φ, such that u = −∇Φ, the

zero-shear-stress boundary condition, is automatically satisfied by the inviscid

assumption. For incompressible potential flows, Φ satisfies Laplace’s equation

which can be solved with the method of separation of variables [12], and upon

differentiation yields the potential flow velocity fields, ur,PF and uθ,PF , as

presented in eqn. (8) and (9).

3

R

ur,PF (r, θ) = u∞ − 1 + cos(θ) (8)

r

3

1 R

uθ,PF (r, θ) = u∞ 1 + sin(θ) (9)

2 r

The solutions for both simplifying assumptions are simple algebraic expressions

which allow for rapid numerical function evaluation. However, neither of these

solutions are able to capture the flow separation and wake features which are

artefacts arising when Re > 1, resulting in poor flow prediction at the rear of

the bubble and incorrect hydrodynamic force calculation which limits the practical

applicability of these solutions.

The need for statistical modelling arises from an attempt to bridge the gap between

the two extrema cases of the analytical solutions i.e. Re = 0 and Re = ∞,

by accounting for features occurring in the non-idealized flow regimes. Since the

aim of the BCM is rapid functional evaluation, a semi-analytical model, expressed

through algebraic expressions, would be justified.

The two analytical models, associated with creeping and potential flow, is to be

combined to take account of a significant portion of the flow structure. A linear

combination of the two solutions is used, i.e. eqn. (10) and (11), where α1,1 to

α2,2 represents the weighting coefficients obtained through linear least squares

with respect to utrue . For this purpose, only data from the region π/2 ≤ θ ≤ 3π/2

is considered, since this region is best represented by the two solutions, whereas

the wake feature at the rear of the bubble negatively impacts the least squares fit by

attempting to take account of a structure the model is not capable of representing.

uM

r,A = α1,1 ur,CF + α1,2 ur,PF (10)

uM

θ,A = α2,1 uθ,CF + α2,2 uθ,PF (11)

The residuals to these solutions, i.e. eqn. (12) and (13), require statistical

modelling to be accounted for. This has been achieved by considering each of

the residual velocity surfaces, ur,res and uθ,res across the Reynolds operating

range, and identifying regions following simple functional forms. In addition, the

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 281

number . R1 and R2 denote region 1 and region 2.

symmetry across the x = 0 line, was also used to simplify the problem.

ur,res = utrue

r − uM

r,A (12)

uθ,res = utrue

θ − uM

θ,A (13)

The residual of to the radial velocity, was largely focussed in the bubble wake

region. This was captured by identifying a critical angle, θc , corresponding to the

maximum of the local peak at the rear of the bubble, which divides the domain into

two regions, as illustrated in Figure 4. These regions were subsequently captured

through a series of exponential models. The final model for uM r had a total of 17

parameters (α1,i ) and resulted in R2 values between 0.9968 - 0.9999. The residual

of the angular velocity was a maximum at the bubble surface, which decayed into

the radial direction. This surface was approximated through finding the appropriate

model to represent uθ,res |r=R and fitting the exponential decay of this model

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

282 Advances in Fluid Mechanics IX

r and uθ with varying Reynolds number.

Re uM

r : MSE uM

r :R

2

uM

θ : MSE: uM

θ :R

2

15 7.2915e-08 0.9989 2.9301e-07 0.9978

75 3.9737e-07 0.9998 1.7457e-06 0.9995

135 7.0539e-07 0.9999 3.9877e-06 0.9997

195 1.2357e-06 0.9999 6.9260e-06 0.9997

270 2.2273e-06 0.9999 1.1348e-05 0.9998

to 0.9998, with 11 parameters (α2,i ). (Please note that due to space restrictions it

is not possible to expand on the residual models i.e. uM M

r,res and uθ,res . They will

however be discussed in more detail in an upcoming paper by the authors.)

The overall results are presented in Table 1. It is seen that the accuracy of

the approximation increases with Reynolds number. This is due to the deviations

from the potential flow solution, mainly consisting of the wake feature, becoming

concentrated in a smaller region as Re increases, thereby allowing for a more

accurate fit. At the lower range the wake is more spatially spread out, and whilst

the uM M

r,A and uθ,A approximations becomes better as the creeping flow solution is

approached at Re = 0.1, the uM M

r,res and uθ,res models are not as accurate at lower

Reynolds numbers. Response surface strategies have also been evaluated for the

same data, however they resulted in higher MSE’s with more parameters [14].

To ensure a smooth response of the first stage parameters, αi,j , with respect

to Reynolds number, each first stage fit was carried out using the parameter

estimations from the previous fit as initial guesses, thereby warm starting the

optimization e.g. the parameter estimates for the Re = 270 surface is used as

initial guesses for the Re = 255 surface. Carrying out the procedure from high to

low Reynolds numbers provided the most well behaved curves. The dependence of

the α s w.r.t. Re was found to be described by first and second order polynomial

and exponential models. A typical example of the behaviour of the α s is given in

Figure 6.

The final parameters to the BCM was found by optimizing all of the ζ

parameters with respect to the original CFD surfaces i.e. utrue . This results in a

bubble flow field model dependent on the Reynolds number (Re), spatial position

(r and θ) and the stage two fitting parameters (ζ).

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 283

single bubble test case

With the BCM constructed, the next step is to integrate it with a macro flow model

to account for the continuum outside of the BCM regions. For the development

of this model, the Openfoam C++ libraries where used [15]. The BCM regions

are tracked in a Lagrangian framework, and the Navier–Stokes equations solved

on a structured grid for the continuum. A structured grid is used such that the

computational cells falling within the BCM regions can be identified, through

their structured relationship, in a computationally efficient manner. To exclude

the BCM regions from the Eularian flow computations, the linear system of

equations is modified. The corresponding values are set to that of the BCM

and the cells are eliminated from the equations. Furthermore, the edges of the

BCM regions neighbouring the computational cells, act as “internal boundary

conditions”, thereby closing the problem.

As a pilot test to the BCM approach, the case of a single bubble rising is

considered with a fixed Reynolds number as illustrated in Figure 7. The resulting

flow field in the continuum qualitatively follows experimental and numerical

modelling trends [16, 17]. A rigorous analysis and validation of the results

however, remain to be completed at this stage. Furthermore, the equation of motion

for the BCM case needs to be developed.

4 Conclusion

A novel bubble column modelling approach is proposed, with the goal to offer

bubble scale flow information at reduced computational expense. This is aimed

to be achieved through the introduction of an algebraic flow model which

approximates the flow field around individual bubbles, i.e. the BCM, in contrast

to having to solve a set of non-linear PDE’s. The BCM was constructed through a

two stage fitting process, where the parameters are correlated with CFD generated

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

284 Advances in Fluid Mechanics IX

Lagrange framework following the BCM approach. The bubble is

simulated as Re = 120.

data, for the analogues situation of flow over a fixed fluid sphere. The combination

of the creeping and potential flow analytical solutions provided a good starting

point by taking account of a large portion of the flow field. The non-linear

deviations occurring outside of their idealized operating regimes however, required

the introduction of statistical models. The combination of these models resulted in

a semi-analytical model capable of approximating the flow field very accurately in

the range examined. Furthermore, the cross correlation of the stage one parameters

with Reynolds number proved to be successful with simple polynomial and

exponential functions, thereby arriving at the BCM.

The BCM has been integrated with a “macro Eularian” flow model, whereby

the BCM regions are tracked in a Lagrangian framework. This has been performed

successfully with a pilot test case, where a single bubble rising is simulated with

a fixed Reynolds number. Rigorous validation remains outstanding, however, the

results qualitatively appear realistic and illustrates the potential of the approach.

References

[1] Brenner, C.E., Fundamentals of Multiphase Flow. Cambridge University

Press, 2005.

[2] Michaelides, E.E., Particles, Bubbles & Drops: Their Motion, Heat and Mass

Transfer. World Scientific, 2006.

[3] Joshi, J.B., Computational flow modelling and design of bubble column

reactors. Chemical Engineering Science, 56, pp. 5893–5933, 2001.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 285

[4] Sokolichin, A., Eigenberger, G. & Lapin, A., Simulation of buoyancy driven

bubbly flow: Established simplifications and open questions. American

Institute of Chemical Engineers, 50, pp. 24–45, 2004.

[5] Magnaudet, J. & Eames, I., The motion of high-Reynolds-number bubbles

in inhomogeneous flows. Annual Review: Fluid Mechanics, 32, pp. 659–708,

2000.

[6] Batchelor, G.K., An Introduction to Fluid Dynamics. Cambridge University

Press, 1967.

[7] Tomiyama, A., Celata, G.P., Hosokawa, S. & Yoshida, S., Terminal velocity

of single bubbles in surface tension force dominant regime. International

Journal of Multiphase Flow, 28, pp. 1497–1519, 2002.

[8] Clift, R., Grace, J.R. & Weber, M.E., Bubbles, Drops and Particles.

Academic Press, 1978.

[9] Pantankar, S.V., Numerical Heat Transfer and Fluid Flow. Taylor & Francis,

1980.

[10] Haberman, W.L. & Morton, R.K. David Taylor Model Basin Rep, 802, 1953.

[11] Stokes, G.G., On the effect of the internal friction of fluids on the motion

of a pendulum. Transactions of the Cambridge Philosophical Society, 9, pp.

8–106, 1851.

[12] Slattery, J.C., Advanced Transport Phenomena. Cambridge University Press,

1999.

[13] Bird, R.B., Steward, W.E. & Lightfoot, E.N., Transport Phenomena: second

edition. John Wiley & Sons Inc, 2002.

[14] Coetzee, W., Coetzer, R.J.L. & Rawatlal, R., An analysis of response surface

strategies in developing statistical models for application in bubble column

modelling. submitted to: Computers & Chemical Engineering, 2010.

[15] OpenCFD, http://www.openfoam.com/, OpenFoam - The Open Source CFD

Toolbox, 2009.

[16] Ellingsen, K. & Risso, F., On the rise of an ellipsoidal bubble in water:

oscillatory paths and liquid-induced velocity. Journal of Fluid Mechanics,

440, pp. 335–268, 2001.

[17] Deen, N.G., van Sint Annaland, M. & Kuipers, J.A.M., Multi-scale modeling

of dispersed gas-liquid two-phase flow. Chemical Engineering Science, 59,

pp. 1853–1861, 2004.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

This page intentionally left blank

www.engbookspdf.com

Advances in Fluid Mechanics IX 287

of the churning power losses in an industrial

planetary speed reducer

F. Concli & C. Gorla

Politecnico di Milano, Dipartimento di Meccanica, Milan, Italy

Abstract

Advantages of planetary speed reducers are well known and due to their compact

design and power density they are suitable for a wide range of applications.

Efficiency is becoming more and more of a main concern in the design of power

transmissions and the demand for high efficiency gearboxes is continuously

increasing. For this reason it is important to have some models in order to

quantify the power losses of those already during the design stage. Some

theoretical or semi-empirical models that allow us to estimate losses like those of

bearings, of seals, of gear meshing (due to sliding) and, for ordinary gears, those

of churning are available in literature. In the case of planetary speed reducers,

where the motion of the planet carrier causes a rotatory motion of the planets

around the axis of the gearbox, the oil splash lubrication is an important source

of losses. This report introduces a multiphase CFD model for the prediction of

the churning losses characteristic of planetary gears. The analysis has been

carried out by means of an unsteady Volume of Fluid (VOF) model and

implemented on a commercial software (Fluent). The whole geometry of the

speed reducer and many operating conditions like the rotational speed, the oil

level and the operating temperature have been taken into account. Moreover the

results of an experimental testing campaign on an especially designed gearbox

are presented and compared with the computation ones in order to validate the

model. The two approaches give results in good agreement.

Keywords: planetary speed reducers, churning losses, multiphase flow

simulation, VOF.

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120261

www.engbookspdf.com

288 Advances in Fluid Mechanics IX

1 Introduction

In the last years, efficiency is becoming more and more a main concern also in

the design of power transmissions and the demand for high efficiency gearboxes

is continuously increasing. Just having appropriate models to predict the final

efficiency of the speed reducer [1] it is possible to take the correct choices since

the design step, avoiding waste of time and money. Some models for the

estimation of the different sources of losses like those of bearings [2], of seals

[3], of gear meshing (due to sliding) [3] and, for ordinary gears, those of

churning [4] already exist. What is still missing, is an appropriate model for the

prediction of the churning losses in planetary speed reducers. In this kind of

gearing, in fact, the motion of the planetary gears due to the planet carrier

rotation is an additional source of losses.

Aim of this study is to provide a model for the correct estimation of this kind

of losses and, consequently, for the correct estimation of the efficiency of the

whole transmission. The model has been performed by mean of a computational

fluid dynamic (CFD) analysis that simulates the behaviour of the air-oil lubricant

mixture.

To validate the numerical model, an industrial planetary speed reducer has

been conveniently modified in order to be able to measure the churning losses

alone caused by the motion of the planetary gears. This gearbox has been tested

on an especially designed test rig.

2 Problem definition

Figure 1 shows an example of a planetary gearing: it can be seen the sun gear,

the planetary gears, the planet carrier and the external crown.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 289

In the common configuration, the power flows from the sun gear shaft to the

planet carrier shaft. The planetary gears have therefore two rotatory motion

components: the first one is a rotation around their axis and the second one is a

motion with a circular path around the gearbox axis due to the planet carrier

rotation on which they are mounted.

The planetary speed reducers are generally oil splash lubricated and it is just

the interaction between the rotating elements and the lubricant air-oil mixture the

source of losses investigated by the presented model.

The influence parameters are the oil level and its properties, such as the

viscosity and the density (functions of the temperature), the geometry and, of

course, the rotational speed. The transmitted torque influences this kind of losses

indirectly: increasing the transmitted torque means more load dependent power

losses (like the meshing ones and part of the bearing ones) and, consequently, a

higher regime temperature of the lubricant.

3 Geometry modifications

The initial geometry of the analysed speed reducer is shown in figure 2.

The speed reducer is composed by a sun gear, 3 planetary gears and a “double

disk” planet carrier. Both, the input and the output shaft, are mounted by mean of

2 bearings. Two contact seal are also present to avoid leakage of lubricant. In

order to evaluate the churning power losses related to the motion of the planetary

gears around the gearbox axis due to the rotation of the planer carrier only, the

originary geometry of the speed reducer has been modified. The teeth of the

external crown have been removed by machining and the sun gear has been

substituted with a smaller and cylindrical (without teeth) component. The aim of

these modifications is to prevent the engaging of the gears, avoiding sliding

losses and churning losses due to the rotation of the gears around their axis. This

modified geometry has been used both for the numerical model and the

experimental testing campaign.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

290 Advances in Fluid Mechanics IX

4 Model set up

4.1 Geometry and mesh

The domain for the CFD analysis is the internal free volume of the speed reducer

after the above described geometry modifications. Figure 3 shows this volume

marked in grey.

is marked in yellow.

The computational domain for the CFD analysis has been modelled by means

of 3D cad software and discretized with a swept mesh. This meshing technique

consists in creating a mesh on one side of the region, known as the source side,

and then copying the nodes of that mesh, one element layer at a time, until the

final side, known as the target side, is reached.

The whole model has been discretized with about 500000 hexahedral cells.

This kind of elements allows a larger aspect ratio compared with the tetrahedral

cells in which it will invariably affects the skewness of the cell, which is

undesirable as it may impede accuracy and convergence.

A VOF multiphase approach has been chosen for the analysis. The VOF

formulation relies on the fact that two or more fluids (or phases) are not

interpenetrating. The fields for all variables and properties are shared by the

phases and represent volume-averaged values, as long as the volume fraction of

each of the phases is known at each location. Thus the variables and properties in

any given cell are either purely representative of one of the phases, or

representative of a mixture of the phases, depending upon the volume fraction

values. The tracking of the interface between the phases is accomplished by the

solution of a continuity equation for the volume fraction of one of the phases

∙ ∑ (1)

where is the density of the phase, the volume fraction of that phase,

and the mass transfer from phase to phase and vice versa

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 291

explicit scheme discretisation in order to avoid the numerical diffusion. That

means solving

∑ , ∑ (2)

∆

where 1 and are the indexes for the new time step and the preious one,

, is the face value of the volume fraction computed with the Geo-

Reconstruction scheme, is the volume of the cell and is the volume flux on

the face. The Geo-Reconstruction approach is an accurate scheme that assumes

that the interface between two fluids has a linear slope within each cell, and uses

this linear shape for the calculation of the advection of fluid through the cell

faces. This scheme avoids the numerical diffusion but needs an accurate grid.

The properties appearing in the above transport equations are determined by the

presence of the component phases in each control volume. In a two-phase

system, if the phases are represented, for example, by the subscripts air and oil,

and if the volume fraction of the second of these is being tracked, the density in

each cell is given by

1 (3)

All other properties are calculated in the same manner.

The two phases properties are summarized in table 1.

Air 1.225 1.7894E-05

Lubricant Oil 1041 0.2082

A single momentum equation is solved throughout the domain, and the

resulting velocity field is shared among the phases

∙ ∙ (4)

velocity differences exist between the phases, the accuracy of the velocities

computed near the interface can be adversely affected. The energy equation si

also shared among the phases

∙ ∙ (5)

variables. As with the velocity field, the accuracy of the temperature near the

interface is limited in cases where large temperature differences exist between

the phases.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

292 Advances in Fluid Mechanics IX

suggested for flows in closed domains. This algorithm uses a relationship

between velocity and pressure corrections to enforce mass conservation and to

obtain the pressure field.

In order to reproduce the operating conditions of the speed reducer, a rigid

motion of the mesh has been applied. The motion has been defined by mean of

an UdF (User defined Function). All the boundaries has been set to no slip walls:

the internal boundaries, corresponding to the planet carrier and planetary gears

surfaces, rotate together with the mesh (marked in dark grey in the figure 4)

while the other boundaries, corresponding to the external crown and the case of

the gearbox (marked in black in the figure 4), are stationary in the absolute

reference frame.

a) b)

A-A fig. 3) b) 3D representation of the computational domain.

The time step for the transient analysis has been evaluated as

,

∆ (6)

where , is the volume of the smallest cell in the computational domain

and is the velocity scale of the problem.

The purpose of the simulations is to analyze the resistant torque on the planet

carrier shaft. This moment is calculated with a surface integral on the moving

walls with respect to the gearbox axis and it is composed by two parts: the first

given by pressure and the second by the viscous effects. Some simulations have

been computed with different combinations of oil level, operating temperature

and rotational speed. Table 2 shows the combinations of parameters for each

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 293

planet carrier in rpm and is the oil level in mm measured from the gearbox axis

(positive if higher than the axis, negative if lower).

40 20 500

40 0 500

40 0 1500

40 20 1000

40 20 1500

40 0 1000

40 -20 500

40 -20 1000

40 -20 1500

20 0 1000

65 0 1000

The analysis has been stopped after the resistant torque had no more sensible

fluctuations and has stabilized. By multiplying the mean value of the torque by

the imposed rotational speed, it is possible to determine the churning power

losses.

a) b)

volume fraction for the oil phase in the mid-section.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

294 Advances in Fluid Mechanics IX

5 Experimental tests

In order to provide a validation of the numerical model, a real industrial

planetary speed reducer has been modified as already described and tested. After

the modifications, the input shaft and the output shaft are completely uncoupled.

That allows us to move the planet carrier and reproduce the condition in which

the meshing losses and the churning losses due to the rotations of the gears

around their axis are avoided.

A schematic layout of the test rig is shown in figure 6. The speed reducer is

fixed to the ground by mean of a specially designed structure. The planet carrier

shaft is connected to a HBM T12 torque meter with a telemetric transmission of

the signal by mean of a double cardan shaft in order to avoid bending. The

torque meter, in turn, is connected to a 35KW DC motor by means of a coupling.

A transparent pipe allows the monitoring of the actual oil level. The operating

temperature can be imposed by a special heating system (insulated chamber with

a heating source) and measured by mean of a thermocouple. The rotational speed

can be controlled by the motor PLC (programmable logic controller). For each of

the 3 oil levels, 4 different temperatures have been tested. For all this coupled

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 295

operating conditions (oil level + temperature) the power losses have been

measured in a rotational speed range between 100 rpm and 1500 rpm.

6 Results

Before comparing the experimental results with the computation ones, it is

necessary to subtract the seal losses and the bearing losses from the measured

values. The torque meter, in fact, measures the input power of the reducer. A

little fraction of this power is dissipated by the two internal bearings of the planet

carrier shaft and by its seal. This part of losses can be easily calculated by mean

of some proved models [2, 3] as function of the testing conditions.

Figure 8 to 13 show both the experimental and the computation results for the

different oil levels and as a function of temperature and rotational speed.

Power losses L=+20mm

300

ω=500rpm (exp)

250

ω=1000rpm (exp)

200

ω=1500rpm (exp)

P [W]

150

ω=500rpm (CFD)

100

ω=1000rpm (CFD)

50

ω=1500rpm (CFD)

0

0.0 50.0 100.0

T [°C]

Power losses L=0mm

250

ω=500rpm (exp)

200 ω=1000rpm (exp)

150 ω=1500rpm (exp)

P [W]

100 ω=500rpm (CFD)

ω=1000rpm (CFD)

50

ω=1500rpm (CFD)

0

0.0 50.0 100.0

T [°C]

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

296 Advances in Fluid Mechanics IX

Power losses L=‐20mm

150 ω=500rpm (exp)

125 ω=1000rpm (exp)

100 ω=1500rpm (exp)

P [W]

75 ω=500rpm (CFD)

50 ω=1000rpm (CFD)

25 ω=1500rpm (CFD)

0

0.0 50.0 100.0

T [°C]

Power losses (L=20mm)

300

T=20°C (exp)

250

200 T=45°C (exp)

P [W]

150 T=65°C (exp)

100 T=85°C (exp)

50 T=40°C (CFD)

0

0 500 1000 1500 2000

[rpm]

Power losses (L=0mm)

300

T=20°C (exp)

200 T=45°C (exp)

P [W]

T=65°C (exp)

100 T=85°C (exp)

T=40°C (CFD)

0

0 500 1000 1500 2000

[rpm]

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 297

Power losses (L=‐20mm)

175

150 T=20°C (exp)

125 T=45°C (exp)

P [W]

100 T=65°C (exp)

75

T=85°C (exp)

50

T=40°C (CFD)

25

0

0 500 1000 1500 2000

[rpm]

Due to the high computational time needed for the solution of the numerical

models, not all the experimentally tested conditions have been also computed

with the CFD analysis.

It can be seen that the numerical results are in good agreement with the

numerical ones.

7 Conclusions

As reliable models to predict the churning losses of the planetary gears are still

not available, a CFD model has been applied in order to predict this important

component of losses of epicycloid gear reducers, starting from the geometry and

the operating conditions. The results of the model are well supported by the

experiments (except for extremely low temperatures). The losses increase, as

expected, with static lubricant level and rotational speed and decrease with the

temperature. The decrease rate of the power losses with the temperature is very

high for low temperature and decreases with the temperature growth.

The increase rate of the losses with the rotational speed is more than linear.

Future development is the investigation of other influence parameters like the

oil type and the geometry of the reducer.

References

[1] Concli, F., Gorla, C., Arigoni, R., Cognigni, E., Musolesi, M., Planetary

Speed Reducers: Efficiency, Backlash, Stiffness, International conference on

gears, Munich 2010.

[2] General Catalogue SKF –– SKF Group, December 2006.

[3] Niemann, G., Winter, H., Maschinenelemente – Band 2: Getriebe Allgemein,

Zahnradgetriebe – Grundlagen, Stirnradgetriebe – 2. Auflage , Springer,

Berlin 2003.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

298 Advances in Fluid Mechanics IX

[5] Patankar, S.V., Numerical heat transfer and fluid flow, Taylor&Francis,

USA 1980.

[6] Versteeg, H.K., Malalasekera, W., An introduction to computational fluid

dynamics – The finite volume method, Longman Group, London 1995.

[7] Comini, G., Fondamenti di termofluidodinamica computazionale,

SGEditoriali, Padova 2004.

[8] Concli, F., Gorla, C., Arigoni, R., Musolesi, M., Riduttori di precisione a

gioco ridotto ed alta efficienza, Organi di trasmissione – febbraio 2011,

Tecniche Nuove, Milano 2011.

[9] Csobàn, A., Kozma, M., Influence of the Oil Churning, the Bearing and the

Tooth Friction Losses on the Efficiency of Planetary Gears, Journal of

Mechanichal Engineering 56(2010)4, pp.231-238.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Section 4

Environmental

fluid mechanics

www.engbookspdf.com

This page intentionally left blank

www.engbookspdf.com

Advances in Fluid Mechanics IX 301

multi-layer passive flux meter measurements

H. Klammler1,2,3, K. Hatfield2,3, J. Luz1, M. Annable3,4,

M. Newman2,3, J. Cho3,4, A. Peacock5, V. Stucker6,

J. Ranville6 & C. Clark II7

1

Department of Environmental Science and Sustainable Development,

Federal University of Bahia, Brazil

2

Department of Civil and Coastal Engineering,

University of Florida, USA

3

Inter-Disciplinary Program in Hydrologic Sciences,

University of Florida, USA

4

Department of Environmental Engineering Sciences,

University of Florida, USA

5

Microbial Insights, Inc., Rockford, USA

6

Department of Chemistry and Geochemistry,

Colorado School of Mines, USA

7

Department of Civil and Environmental Engineering,

Florida A&M University, USA

Abstract

The passive flux meter (PFM) enables the measurement of cumulative water and

contaminant mass fluxes in porous aquifers. It consists of a sorbent material,

which is installed in a monitoring well to intercept groundwater flow. Tracer

losses and contaminant retention on the sorbent are used to estimate water and

contaminant mass fluxes through the device. In the multi-layer PFM different

(sorbent) materials are used in an annulus (layer-type) configuration. This allows

leached tracers inside the PFM (no tracer release into aquifer) to be retained and

facilitates simultaneous deployment of different sorbent types in a single device.

In order to estimate undisturbed ambient fluxes in the aquifer, measurements

need to be corrected for flow convergence or divergence induced by the well and

PFM components. We make use of an analytical solution to the potential flow

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120271

www.engbookspdf.com

302 Advances in Fluid Mechanics IX

hydraulic conductivities. A flow convergence factor is defined as a function of

PFM ring conductivities and radii, where tracer elution and contaminant sorption

may occur in arbitrary layers. The results are used for calibration of convergence

factors of a multi-layer PFM to laboratory sand box experiments.

Keywords: aquifer, groundwater, plume, flow convergence, sand box.

1 Introduction

Groundwater contamination is recognized as a dangerous threat to ecosystems

and human drinking water supplies. Besides contaminant concentrations (mass

per volume), contaminant mass fluxes (mass per cross sectional area per time)

have been used more recently as relevant measures for contaminant source

identification, risk assessment, decision making and remediation performance

control (ITRC [1]). Currently, three fundamental approaches are available for

measuring contaminant fluxes: (1) Multi-level sampling (MLS; Einarson and

Mackay [2]), which is based on separate measurements of contaminant

concentrations and water fluxes for subsequent multiplication to obtain

contaminant fluxes. (2) Integral pump tests (IPT; Bockelmann et al. [3]), which

extract contaminated groundwater from the aquifer through pumping from a well

and monitor contaminant concentrations at the well head over time. (3) Passive

flux meter measurements (PFM; Hatfield et al. [4], Annable et al. [5]) based on

the installation of sorbent materials in observation wells, where the sorbents

initially contain known amounts of resident tracers. From detected tracer losses

from a sorbent and contaminant masses sorbed onto a sorbent through laboratory

analyses, cumulative (i.e., time integrated or averaged over the period of

installation) water and contaminant fluxes may be obtained simultaneously as

depth profiles along the well.

PFMs have typically been deployed in observation wells as self-contained

units consisting of a single sorbent material, which acts as both a leaching tracer

reservoir and a contaminant trap. As a consequence, it has to be assured that (1)

the sorption properties of the sorbent material are appropriate for both tracer(s)

and target contaminants, and (2) that the chemical properties of the tracer(s) are

such that tracer release (even though minimal) into the aquifer does not cause

legal or environmental problems. In an effort to circumvent these two issues

(e.g., for measuring water and contaminant fluxes at Rifle, CO, USA) a multi-

layer PFM has been developed and tested, which consists of multiple concentric

rings (layers) of different materials (which may be sorbents or not).

Figure 1 compares the two different PFM configurations, where figure 1a

shows a single sorbent installed inside a well screen, while figure 1b illustrates

the annular composition of a multi-layer PFM installed in a screened well. From

the periphery towards the center ki [L/T] and ri [L] denote the hydraulic

conductivities and outer radii of the different rings. k0 corresponds to the aquifer,

k1 is the well screen, k2 an outer sorbent layer for contaminant sorption (e.g.,

Lewatit resin for uranium), k3 an intermediate sorbent layer (e.g., granular

activated carbon (GAC) for alcoholic tracers) to retain tracers eluded from an

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 303

inner sorbent layer of k5 (e.g., also consisting of GAC for alcoholic tracers).

Between the two GAC layers, there is a thin perforated stainless steel pipe of

conductivity k4, which serves to separate the inner GAC from the outer one for

installation and laboratory analysis. The center circle of radius r6 is an

impermeable pipe for physical stabilization and water evacuation during PFM

(a) (b)

r5 r3

r6

r2 k5 r2

k2 k3

k2

r1 r1

k1 k1

k0 k0

Figure 1: Horizontal cross sections of well screens and (a) a PFM consisting

of a single sorbent and (b) a multi-layer PFM as used in laboratory

testing. Bold black circle is of outer radius r4 and conductivity k4.

with k1 and k4 to be determined from laboratory box experiments.

Material / purpose

[-] [m/day] [cm]

0 33 infinite Aquifer (sand)

1 k1 5.7 Well screen (slotted PVC pipe)

2 250 5.1 Lewatit resin for uranium sorption

3 350 3.8 GAC for tracer retention

4 k4 2 Perforated pipe for separation

5 350 1.9 GAC for tracer elution

6 0 0.8 Impermeable center pipe

PFM configuration used in laboratory sand box experiments for determination of

k1 and k4 and subsequent deployment for measuring water and uranium fluxes at

the uranium field site in Rifle.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

304 Advances in Fluid Mechanics IX

Similar to the simple PFM, the multi-layer PFM provides data in terms of

tracer losses from sorbent ring 5 and contaminant masses retained from sorbent

ring 2. Following the method of Hatfield et al. [4] the apparent flux qPFM,i [L/T]

through the i-th ring of a PFM may be found as

q PFM ,i

1 m r

ri i ri 21 i Ri

2

(1)

2ri t

where mri [-] is the relative mass of a tracer remaining (with respect to the initial

mass of that tracer) after time of exposure t [T] in the i-th ring of relative water

content θi [-] and retardation factor Ri [-]. The attribute “apparent” indicates that

flow is generally not uniform inside a multi-layer PFM and qPFM,i is to be

understood as a discharge per unit transect area of the i-th ring perpendicular to

incident flow direction. In analogy, an apparent contaminant mass flux JPFM,i

[M/(TL2)] trough the i-th ring may be obtained from

M si

J PFM ,i (2)

2ri bt

where Msi [M] is the mass of contaminant sorbed in the i-th layer over a PFM

interval of length b [L]. Eqn (1) is valid as long as none of the stream tubes

through the i-th layer are completely cleared (empty) of tracer, while eqn (2) is

valid as long as none of the contaminant previously sorbed onto the i-th layer is

again released from it. As a consequence, within their ranges of validity eqns (1)

and (2) do not depend on the properties (e.g., non-uniformity) of the flow field in

the respective rings. Note that for ri+1 = 0 (i.e., the i-th layer is the center circle)

eqn (1) reduces to eqn (18) of Hatfield et al. [4], whose coefficient of 1.67

appears as π/2 ≈ 1.57 here. This is a consequence of the range of validity

stipulated (an analogous observation applies to eqn (2)).

However, due to the more complex configuration of the multi-layer PFM and

flow refraction between layers of different conductivities, an assessment of

undisturbed ambient water and contaminant fluxes in the aquifer is not straight-

forward. Klammler et al. [6] present an analytical solution to the potential flow

problem through the multi-layer PFM and they develop flow convergence factors

for estimation of undisturbed (uniform) ambient fluxes. These factors, however,

are with respect to the inner-most ring (center circle) only and are not applicable

to other rings. The present work generalizes the convergence factors of

Klammler et al. [6] to arbitrary layers in a multi-layered PFM configuration and

uses the result for calibration of unknown parameters (well screen and perforated

pipe conductivities k1 and k4) through a laboratory sand box experiment, such

that they may be applied to the multi-layer PFM deployment at Rifle for

measuring water and uranium fluxes.

The solution of Klammler et al. [6] is based on potential flow through porous

media (Strack [7]). It makes use of a flow field analogy between uniform flow

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 305

flow disturbed by an impermeable or infinitely permeable cylinder. Its

application is best illustrated by a numerical example, for which we use the flow

domain of figure 1b (table 1) with values of k1 = 2.3 m/d and k4 = 3.2 m/d. This

leads to the flow field (stream lines) of figure 2 with the stream function Ψi

[L3/T] in the i-th ring given by

a2

i qi rb1 i2 sin (3)

r

where r [L] and γ [-] are the radial and angular coordinates, respectively, b [L] is

the thickness of the flow domain (length of PFM interval in eqn (2)) and the

parameters ai [L] and qi [L/T] are given in table 2 and obtained as follows:

Starting with an initial value of a6 = 0 in table 2, the columns of kia [L/T] and ai

are populated from bottom up by alternately applying eqns (4) and (5).

Subsequently, the column of qi is populated from top down by consecutive use of

eqn (6). Most parameters in table 2 are auxiliary variables without direct physical

equivalences (some of them complex/ imaginary). Exceptions are q0 (undisturbed

flux in the aquifer; assumed uniform), k6a (equal to conductivity k6 of inner ring),

k0a (equal to aquifer conductivity k0) and q6 (specific discharge in inner ring).

Since k6 = 0, it is further seen that a5 is equal to the center pipe radius r6.

Following this example, the step-wise solution scheme of eqns (3) through (6) is

generally applicable to an arbitrary number of layers.

2

a

1 i

k ia k i ri (4)

2

a

1 i

ri

kia

1

ki 1

ai1 ri (5)

k

1 ia

ki1

ki

1

kia

qi qi 1 (6)

k

1 i1

kia

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

306 Advances in Fluid Mechanics IX

Figure 2: Flow field solution depicted as stream lines (lines of constant Ψi)

for flow domain of figure 1b (table 1).

multi-layer PFM of figure 1b for k1 = 2.3 m/d and k4 = 3.2 m/d.

j 1 is the imaginary unit.

0 2.94 33 q0/q0 = 1 1

1 5.05j 19.12 0.41 0.73

2 0.79 238.38 0.83 0.81

3 1.73 229.46 1.01 0.80

4 1.88j 49.78 0.13 0.25

5 0.80 244.59 0.32 0.26

6 a6 = 0 0 0 0

Using the solution of Ψi from eqn (3) for an arbitrary number of layers, the

apparent flux qPFM,i [L/T] through the i-th ring of eqn (1) may be found. It is

equal to the difference of the stream function between the lateral-most points

(e.g., for i = 2 points A and B in figure 2) of the ring divided by the cross

sectional area perpendicular to flow.

i (ri , ) i (ri , ) i (ri , )

2 q 1 ai

2

q PFM ,i 2 2 (7)

i

2ri b ri b ri 2

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 307

In the undisturbed aquifer the flow is assumed to be of uniform flux q0, such

that a flow convergence (or divergence) factor αi [-] for the i-th ring may be

defined by

q PFM ,i qi ai2

i 1 2 (8)

q0 q0 ri

In other words, the flow convergence factor expresses how much flow crosses

the i-th ring of a multi-layer PFM with respect to the ambient flow through a ring

of the same size, if the aquifer was not disturbed by the well and PFM

components. A value of αi larger than one may be viewed as a situation of flow

convergence, while a value smaller than one reflects a condition of flow

divergence. For a single layer PFM, flow convergence occurs when the PFM

sorbent is more permeable than the aquifer and flow divergence occurs when the

contrary is true. For multi-layer PFMs, however, αi ≥ 1 and αi ≤ 1 may occur

simultaneously in different layers of a single device depending on the sequence

of ki. Returning to the example of figure 1b and using values of ri from table 1

with values of qi/q0 and ai from table 2, the values of αi as given in the last

column of table 2 are directly obtained from eqn (8) (note hereby that some

values of ai are imaginary).

The flow convergence factor αi from eqn (8) is a generalization over that of

Klammler et al. [6], because it is applicable to any layer. If i is equal to the total

number of layers present, such that αi applies to the center circle of the flow

domain as shown in figure 1, for example, then eqn (8) becomes equal to αi from

Klammler et al. [6]. Moreover, for i = 1, 2 and 3 (and an arbitrary total number

of layers), eqns (3), (12) and (13) of Klammler et al. [6] may be written in a

generalized form by using k1a, k2a and k3a from eqn (4) instead of k1, k2, and k3

resulting in

2 (9)

1

k

1 0

k1a

4

2 2

(10)

k0 k k0 k r2

1 1 1 1 1 1

k1 k 2 a k1 k 2 a r1

8

3 2 2 2

k k k k k k r k k k r k k k r

1 0 1 1 1 2 1 0 1 1 1 2 2 1 0 1 1 1 2 3 1 0 1 1 1 2 3

k1 k 2 k3a k1 k 2 k3a r1 k1 k 2 k3a r2 k1 k 2 k3a r1

(11)

For i = 4, the result of Appendix B in Klammler et al. [6] is incomplete and

after substituting k4 by k4a it should be

16

4 (12)

A B C D E F G H

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

308 Advances in Fluid Mechanics IX

with

k k k k

A 1 0 1 1 1 2 1 3

k1 k2 k3 k4a

2

k k k k r

B 1 0 1 1 1 2 1 3 2

k1 k 2 k 3 k 4 a r1

2

k k k k r

C 1 0 1 1 1 2 1 3 3

k1 k 2 k3 k 4 a r2

2

k k k k r

D 1 0 1 1 1 2 1 3 3

k1 k2 k3 k 4 a r1

2

k k k k r

E 1 0 1 1 1 2 1 3 4

k1 k 2 k 3 k 4 a r3

2

k k k k r

F 1 0 1 1 1 2 1 3 4

k1 k2 k3 k 4 a r2

2

k k k k r

G 1 0 1 1 1 2 1 3 4

k1 k2 k3 k 4 a r1

2

k k k k r r

H 1 0 1 1 1 2 1 3 2 4

k1 k 2 k 3 k 4 a r1 r3

(13)

Note that eqns (9) through (13) are equivalent formulations to the step-wise

procedure of eqns (4) through (6). While closed form expressions of αi for i > 4

may be derived, they become increasingly lengthy and computational

implementation of eqns (4) through (6) (e.g., in a spreadsheet) may be more

convenient.

Not immediately obvious from eqns (9) through (13) and more easily

verifiable using eqns (5), (6) and (8) is that the ratio αi/αi-1 is only a function of

km and rm, where m ≥ i – 1. More intuitively, this means that flow refraction

between two adjacent layers does not depend on the radii and conductivities of

any outside layers or the aquifer. This may be convenient for comparing flux

estimates from different layers independent of the perhaps uncertain aquifer

conductivity k0 and will be explored for estimating k1 and k4 below. However, it

also precludes the possibility of estimating an unknown k0 from two flux

estimates in different layers (as may be attempted in analogy to the method

presented in Klammler et al. [6]).

Assuming that contaminant transport is dominated by advection (i.e.,

contaminant particles travel along the same stream tubes as water particles and

effects of diffusion and dispersion are neglected), qPFM,i and q0 in the first

equality of eqn (8) may be substituted by JPFM,i from eqn (2) and the undisturbed

ambient contaminant mass flux J0 [M/(L2T)], respectively. It is recalled that mri

for estimation of qPFM,i and Msi for estimation of JPFM,i do not have to stem from

the same PFM layer (i.e., index i in eqns (1) and (2) generally takes different

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 309

values). However, if mri and Msi are obtained from the same layer, then the same

flow convergence factor αi applies.

For application of the multi-layer PFM configuration of figure 1b at the uranium

site in Rifle and determination of unknown conductivities k1 and k4, laboratory

sand box experiments are performed. Box size is 39 x 30.5 x 17.9 cm (length L x

width W x height H) and table 1 contains the sand conductivity as well as well

screen and PFM parameters. “Ambient” water fluxes q0 and uranium fluxes J0

through the box are obtained from

Q0

q0 (14)

WH

J 0 q 0 Cu (15)

where Q0 [L3/T] is the independently measured water discharge through the box

and Cu [M/L3] is the uranium concentration in the influent water. Eleven tests of

different durations were run for water flux obtaining estimates qPFM,5 from eqn

(1) for comparison to q0 of eqn (14). In five of these tests uranium was added at

Cu ≈ 200 μg/l for comparison of JPFM,2 from eqn (2) with J0 of eqn (15).

t q0 qPFM,5 α5 J0 JPFM,2 α2

[days] [cm/day] [cm/day] [-] [μg/(cm2day)] [μg/(cm2day)] [-]

3.00 15.95 4.12 0.26

2.87 16.85 4.03 0.24

2.92 16.85 3.44 0.20

N/A

2.99 18.48 4.13 0.22

4.24 17.03 3.84 0.23

4.05 19.44 4.09 0.21

5.92 7.87 2.40 0.30 1.68 1.86 1.11

3.94 11.52 3.64 0.32 2.38 1.80 0.76

1.95 25.02 6.17 0.25 5.05 2.16 0.43

1.51 31.78 7.93 0.25 6.15 2.96 0.48

12.00 3.87 1.68 0.43 0.80 1.05 1.30

Average α5 = 0.26 Average α2 = 0.81

Table 3 summarizes the results of the sand box experiments and indicates

average values of α2 = JPFM,2/J0 = 0.81 and α5 = qPFM,5/q0 = 0.26. At this point we

return to table 2 and revert the previous assumption that the conductivities k1 and

k4 of the well screen and the perforated pipe are known. Instead, the

experimental values of α2 and α5 are used to determine effective values of k1 and

k4 to be used in the interpretation of field deployments. For this purpose,

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

310 Advances in Fluid Mechanics IX

advantage is taken of the previous conclusion that the ratio α5/α2 = 0.32 is not

affected by k1. Thus, k4 may be directly found, which is most conveniently

achieved by systematically varying k4 for an arbitrary value of k1 and observing

the results in terms of α5/α2. Figure 3 shows the outcome of this process and

leads to two possible values of k4 = 3.2 and 18851 m/day. For each of the values

found for k4, the same process is repeated with k1 to reach the required value of

α2 = 0.81 (or equally α5 = 0.26). This is illustrated in figure 4 and shows that

again two values of k1 may be combined with each value of k4, thus resulting in

the four solutions given in the first two rows of table 4.

1

0.9

0.8

k4 = 3.2 m/day k4 = 18851 m/day

0.7

0.6

5/ 2 [-]

0.5

0.4

5/2 = 0.32

0.3

0.2

0.1

0 -5 0 5 10

10 10 10 10

k4 [m/day]

α5/α2 = 0.32.

2 2

k4 = 3.2 m/day k4 = 18851 m/day

1.8 1.8

k = 2.3 m/day k1 = 3450 m/day k = 2.2 m/day

1 1

1.4 1.4

1.2 1.2

2 [-]

2 [-]

1 1

2 = 0.81 2 = 0.81

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 -5 0 5 10

0 -5 0 5 10

10 10 10 10 10 10 10 10

k1 [m/day] k1 [m/day]

value α2 = 0.81.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 311

α2 = 0.81 and α5 = 0.26 from box experiments and resulting αq and

αJ for the Rifle deployment.

k4 3.2 18851

αq 0.20 0.36 0.20 0.37

αJ 0.61 1.13 0.61 1.16

For each of the four solutions pairs, a flow convergence factor αq for water

flux and a flow convergence factor αJ for uranium may be computed for the

deployment conditions at the Rifle site. These conditions are identical to those of

figure 1b, except for the presence of a filter pack of radius 10.2 cm and

conductivity 160 m/day around the well screen and inside an aquifer of

conductivity of approximately 2.5 m/d. Considering these modifications, results

for αq and αJ from application of eqns 4, 5, 6 and 8 are given in the last two rows

of table 4. It may be seen that αq/αJ = α5/α2 = 0.32 remains constant and, more

interestingly, that the solutions are pair wise identical (up to chart reading and

rounding errors). It appears that this is not a coincidence as the same behaviour

may be observed for other hypothetical values of aquifer and filter pack

conductivities and radii. From the remaining two solution, αq = 0.20 and αJ =

0.61 are proposed for use at the Rifle site, since they are associated with a low

value of k1 = 2.3 (and either value of k4). This choice is justified by two related

arguments: (1) From an independent borehole dilution test (no PFM installed) in

the sand box a value of k1 = 87 m/day is estimated. This may be regarded as an

upper bound for k1 as flow in the vicinity of an open borehole is radial. Flow

components in the tangential direction through the well screen are hindered by

the fact that screen slots are not continuous along the circumference of a PVC

screen. (2) As illustrated by the flow field in figure 2, low k1 and k4 cause flow in

the respective rings to be essentially radial, which is in agreement with the

geometric properties of the screen slots and the pipe perforations acting as water

conduits. Large values of k1 and k4, in turn, would lead to a certain degree of

flow short circuiting along these rings, which is considered less plausible,

particularly since the sorbents are granular and tend to settle into the well and

close possible voids along the pipe or screen surfaces.

5 Summary

Based on an existing solution to the potential flow problem of uniform flow

disturbed by an arbitrary number of concentric rings of contrasting

conductivities, flow convergence (or divergence) factors are developed for

interpretation of multi-layer PFM measurements. Flow convergence factors

convert water and contaminant fluxes measured in an arbitrary layer of the PFM

into estimates of respective undisturbed ambient fluxes (i.e., unaffected by the

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

312 Advances in Fluid Mechanics IX

presence of well and PFM). Using the results in combination with laboratory

sand box experiments, effective conductivities of a well screen and another

separation screen between layers are determined. With these conductivities, flow

convergence factors for measuring water and uranium fluxes at a site in Rifle,

CO, USA, are proposed.

Acknowledgements

This research was partially funded by the first author’s fellowship of the Bahia

State Science Foundation (FAPESB; DCR 0001/2009), Brazil, the U.S. National

Science Program (award number 0804134), the Environmental Remediation

Science Program (ERSP), U.S. Department of Energy (grant number DE-FG02-

08ER64585), the U.S. Department of Defense (project number ER0831) under

the Environmental Security Technology Compliance Program (ESTCP) and the

Florida Water Resources Research Center under a grant from the U.S.

Department of the Interior (Grant Number G11AP2007).

References

[1] ITRC, Use and Measurement of Mass Flux and Mass Discharge.

MASSFLUX-1. Interstate Technology & Regulatory Council: Washington,

D.C., 2010.

[2] Einarson, M.D., Mackay, D.M., Predicting impacts of groundwater

contamination. Environmental Science and Technology, 35(3), pp. 66A-

73A, 2001.

[3] Bockelmann, A., Zamfirescu, D., Ptak, T., Grathwohl, P. and Teutsch, G.,

Quantification of mass fluxes and natural attenuation rates at an industrial

site with a limited monitoring network: a case study. Journal of

Contaminant Hydrology, 60(1-2), pp. 97-121, 2003.

[4] Hatfield, K., Annable, M., Cho, J., Rao, P.S.C. and Klammler, H., A direct

passive method for measuring water and contaminant fluxes in porous

media. Journal of Contaminant Hydrology, 75(3-4), pp. 155-181, 2004.

[5] Annable, M., Hatfield, K., Cho, J., Klammler, H., Parker, B., Cherry, J. and

Rao, P.S.C., Field-scale evaluation of the passive flux meter for

simultaneous measurement of groundwater and contaminant fluxes.

Environmental Science and Technology, 39(18), pp. 7194-7201, 2005.

[6] Klammler, H., Hatfield, K., Annable, M., Agyei, E., Parker, B., Cherry, J.

and Rao, P.S.C., General analytical treatment of the flow field relevant to

the interpretation of passive flux meter measurements. Water Resources

Research, 43(4), W04407, 2007.

[7] Strack, O.D.L., Groundwater Mechanics, Prentice Hall: Englewood Cliffs,

N.J., 1989.

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 313

on a sloping bottom in spherical geometry

G. E. Swaters

Applied Mathematics Institute, Department of Mathematical

& Statistical Sciences and Institute for Geophysical Research,

University of Alberta, Canada

Abstract

A steady nonlinear planetary-geostrophic model in spherical coordinates is

presented describing the hemispheric-scale meridional flow of grounded abyssal

currents on a sloping bottom. The model, which corresponds mathematically to a

quasi-linear hyperbolic partial differential equation, can be solved explicitly for a

cross-slope isopycnal field that is grounded (i.e., intersects the bottom on the up

slope and down slope sides). The abyssal currents possess decreasing thickness in

the equatorward direction while maintaining constant meridional volume transport

and exhibit westward intensification as they flow toward the equator.

1 Introduction

Many of the abyssal currents in the oceans associated with the equatorward

motion of deep water masses produced in high latitudes due to atmospheric

cooling, are organized as mesoscale topographically-steered geostrophically-

balanced grounded gravity currents that flow along sloping continental boundaries.

These currents form an important component in the deep “leg” of the meridional

overturning circulation in the oceans. The mesoscale dynamics of these currents

has been described in a series of papers [1–11]. All of these studies have implicitly

assumed either an f or β-plane approximation in which the dynamics is modelled

in a Cartesian coordinate system with the implicit assumption that the horizontal

length scales are not too much larger than the internal deformation radius (on the

order of about 10–100 km in the ocean).

www.witpress.com, ISSN 1743-3533 (on-line)

doi:10.2495/AFM120281

www.engbookspdf.com

314 Advances in Fluid Mechanics IX

hemispheric in scale. This raises the question about the role of planetary sphericity

in determining the large scale kinematic structure of these flows.

The principal purpose of this paper is to present a preliminary report on a

simple, but nevertheless illuminating, reduced-gravity shallow-water model for

the meridional, or equatorward, steady flow of grounded abyssal currents on a

longitudinally-sloping bottom on a rotating sphere. A complete discussion of the

dynamical properties as well as a specific oceanographically-relevant example of

the solution briefly described here will be published elsewhere in a venue more

suitable for research problems in geophysical fluid dynamics.

2 Model derivation

The reduced-gravity shallow-water equations for a stably-stratified abyssal water

mass overlying variable bottom topography on a rotating sphere can be written in

the form

uuλ vuθ u v tan θ

ut + + − − 2Ω v sin θ

a cos θ a a

g

=− (h + hB )λ , (1)

a cos θ

uvλ vvθ v 2 tan θ g

vt + + − + 2Ω u sin θ = − hθ , (2)

a cos θ a a a

1

ht + [(h u)λ + (h v cos θ)θ ] = 0, (3)

a cos θ

where u and v are the zonal (positive eastward) and meridional (positive

northward) velocities, respectively, λ is the longitude (positive eastward) and θ

is the latitude (positive northward), t is time, Ω is the angular frequency associated

with Earth’s rotation (2π rads/day), a is the radius of Earth (about 6300 km),

h (λ, θ, t) is the thickness of the abyssal layer, hB (λ) is the longitudinally-

varying height of the bottom topography above a constant reference depth and

g = g (ρ2 − ρ1 ) /ρ2 > 0 is the (stably-stratified) reduced gravity where ρ1

and ρ2 are the densities of the (infinitely deep and motionless) overlying fluid

and dynamically active abyssal layer, respectively, and g is the gravitational

acceleration (9.81 m/s2 ). Typical oceanographic values for the reduced gravity are

in the range 10−4 −10−2 m/s2 . The dynamic pressure (i.e., the total pressure minus

the hydrostatic pressure) in the abyssal layer is given by p = g ρ1 (h + hB ).

Further analysis is facilitated by introducing the scalings

L a LV

λ= λ, t = t, u = , v = V v,

u

a V a

2ΩV L

(h, hB ) = h, hB , p = 2ΩV Lρ1 p, (4)

g

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

Advances in Fluid Mechanics IX 315

uuλ

ε δ 2 ut + + v uθ − u v tan θ − v sin θ =

cos θ

1

− (h + hB )λ , (5)

cos θ

uvλ

ε vt + + v vθ − v 2 tan θ + u sin θ = −hθ , (6)

cos θ

1

ht + [(h u)λ + (h v cos θ)θ ] = 0, (7)

cos θ

where ε and δ are, respectively, the Rossby number and aspect ratio given by

V L

ε= and δ = ,

2ΩL a

and where the dynamic pressure is given by

p = h + hB .

suggests that

ε 3.4 × 10−4 and δ 1.6 × 10−2 ,

with the additional scalings

a LV 2ΩV L

145 days, 8 × 10−4 m/s and 240 m.

V a g

1

u =− hθ , (8)

sin θ

1

v = (h + hB )λ , (9)

sin θ cos θ

sin2 θ ht + tan θ hBλ hθ − h hλ = hBλ h. (10)

Equations (8)–(10) corresponds to a planetary-geostrophic model in which the

velocities are geostrophically determined but for which order-one dynamic

variations in the thickness in the abyssal layer are permitted, i.e., the thickness field

can intersect the bottom allowing for groundings. It is important to point out that

the model is singular at the equator (θ = 0) and thus cannot be used to described

inter-hemispheric or cross-equatorial flow. The dynamic pressure p = h + hB

forms the geostrophic stream function for the flow as seen in (8) and (9).

www.witpress.com, ISSN 1743-3533 (on-line)

www.engbookspdf.com

316 Advances in Fluid Mechanics IX

We briefly describe the equilibrium solution to the model (8)–(10) in which

h (λ, θ) will be determined by the steady-state quasi-linear hyperbolic equation

h

tan θ hθ − hλ = h. (11)

hBλ

The model (11) can be solved using the method of characteristics with the

boundary condition

h (λ, θ0 ) = h0 (λ) , (12)

where θ0 is a given reference latitude and h0 (λ) is a prescribed abyssal height

profile that varies only in the latitudinal direction.

The solution to (11) subject to (12) can be written in the form

sin θ

h (λ, θ) = h0 (τ ) , (13)

sin θ0

sin θ0 − sin θ

hB (τ ) + h0 (τ ) = hB (λ) . (14)

sin θ0

Given λ and θ, one solves (14) for τ (λ, θ) and then determines h (λ, θ) from (13).

The first thing to note is that the abyssal height monotonically decreases in the

equatorward direction and in the limit as θ → 0, it follows from (13) that

h (λ, θ) → 0.

The characteristics associated with the quasi-linear model (11) are the curves

in (λ, θ)-space along which τ is constant, as determined by (14). Along these

characteristic curves

dθ h (λ) sin θ0

=− B . (15)

dλ τ =constant cos θ h0 (τ )

We are interested in the physical situation where hB (λ) < 0 with θ > 0

corresponding to flow in the northern hemisphere along a topographic slope in

which the depth of the fluid increases as λ increases, i.e., eastward. This is a model

for the equatorward flow of a grounded abyssal water masses along a continental

slope on the western side of an ocean basin. In this situation the characteristics are,

generally speaking, aligned in the southwest to the northeast direction since

hB (λ) sin θ0

− > 0.

cos θ h0 (τ )

In fact, the characteristics correspond to the geostrophic streamlines. Equations

(13) and (14) can be combined into the form

p (λ, θ) ≡ hB (λ) + h (λ, θ) = hB (τ ) + h0 (τ ) . (16)

## Mult mai mult decât documente.

Descoperiți tot ce are Scribd de oferit, inclusiv cărți și cărți audio de la editori majori.

Anulați oricând.